Promiscuous Mutations Activate the Non-Canonical NF-kB Pathway in Multiple Myeloma

aCGH identifies frequent bi-allelic deletions of NF-kB regulatory genes

In order to identify genetic abnormalities in MM we screened 62 patient samples and 46 human myeloma cell lines (HMCLs) originating from 42 individuals on a high density oligonucleotide aCGH platform with 42 896 features at an average genomic interval of ~35 kb (Barrett et al., 2004). We took a reductionist approach to the analysis of the aCGH data by focusing our analysis on bi-allelic/homozygous deletion events, which are intrinsically limited to small genomic regions but are likely to harbor tumor suppressor genes (Figure S1). To ensure that the observed bi-allelic deletions represented genuine events and not noise in the data or large common nucleotide variations (CNV)(Khaja et al., 2006; Redon et al., 2006), we first applied a circular binary segmentation algorithm (CBS) to the data (Olshen et al., 2004). The CBS algorithm provides a very stringent means of identifying bi-allelic deletions as it segregates chromosomal locations into regions with significant adjacent mean copy number changes. Although this reduced the resolution of the analysis from ~35kb to ~70kb, it significantly limits the number of false positive events. This analysis scheme identified 13 bi-allelic deletion events in 11 patients, encompassing 43 genes, and 80 bi-allelic deletions in 36 HMCLs, encompassing 148 genes, that do not represent physiological rearrangements of immunoglobulin loci or aberrations within previously characterized CNV regions (Table S1).

The only recurrent bi-allelic deletion identified by aCGH in our patient samples occurred at 14q32 and encompassed three potential target genes; TRAF3, AMN, and CDC42BPB (Figure 1 & S2). Furthermore, this abnormality was one of only two identified in both our patient and HMCL datasets. We confirmed the bi-allelic deletions by cIg-FISH and the minimally deleted region (MDR) was mapped by PCR in the two HMCLs with bi-allelic deletions to a genomic region of <48.8 kb using sequence tagged site (STS) and fine mapping primer pairs (Figure 1). This fine mapping approach eliminated CDC42BPB as a target gene of the deletion and thus only TRAF3 or AMN remained as potential target genes. The remaining bi-allelic deletion identified in both of our datasets occurred at 11q22. This deletion was often quite large, ranging from 0.14–4.22 Mb in size, however, the MDR defined by aCGH was <109.05 kb in size and encompassed only 2 genes; BIRC2/cIAP1, and BIRC3/cIAP2 (Figure 1). As expected, the presence of a bi-allelic deletion correlated with reduced expression of the predicted target gene (Figure S3).

Interestingly, inactivation of TRAF3, cIAP1/2, and CYLD appears to be a common event in MM, as we identified a number of potential bi-allelic deletions in the publicly available dataset from Carrasco et al., based on a correlation between the aCGH copy number and the expression level (Figure S4)(Carrasco et al., 2006). Combining these two patient datasets did not help to further refine the MDR encompassing TRAF3 or cIAP1/2, however, it did refine the MDR of the bi-allelic deletion we observed at 16q12 to a 72.05 kb region, which includes two genes, CARD15 and CYLD (Figure S4).

Interestingly, 5/13 (38.5%) of the bi-allelic deletion events identified within our patient cohort contained potential target genes (TRAF2, TRAF3, cIAP1/2, and CYLD) associated with the regulation of the NF-kB signaling pathways. Based on these observations we developed the working hypothesis that the frequent dysregulation of NF-kB signaling observed in MM is often the result of genetic abnormalities within the tumor cells.

Combined analysis of GEP and aCGH datasets identifies additional abnormalities within regulators of the NF-kB signaling pathways

Our primary analysis of the aCGH datasets identified inactivation of 5 different genes involved in the NF-kB signaling pathways. To build on these observations we combined our aCGH and previously generated GEP datasets to identify additional abnormalities within genes involved in the NF-kB signaling pathways. A screen of our 125 patient and 44 HMCL GEP cohorts for spiked expression of genes involved in the regulation of NF-kB signaling identified over-expression of MAP3K14/NIK, LTBR, CD40, TNFRSF13B/TACI, and NFKB1 (Figure 2 & S3).

We identified the dysregulated expression of NIK in two HMCLs, L363 and EJM, that have previously been reported to over-express NIK by either a t(17;22) translocation or amplification, respectively (Annunziata et al., 2006). Because of these observations we examined the aCGH datasets for abnormalities of the NIK locus. We confirmed the amplification in EJM, and identified unbalanced rearrangements of NIK in one HMCL, JJN-3, and one patient, which were confirmed by cIg-FISH (Figure 2 & S5).

We noted spiked expression of LTBR and CD40 in both patient and HMCL cohorts. A >40-fold increase in the expression of LTBR was observed in one patient and in the HMCL FR4. For FR4, aCGH showed a significant amplification of the entire 12p chromosome arm, which was confirmed by FISH (Figures 2 & S6). Interestingly, LTBR is one of only six genes on the entire arm with a >40-fold increase in expression. For the one patient with spiked LTBR expression aCGH data was not available; however, FISH identified a single copy gain of the LTBR locus that is inserted at an unknown location (Figure S6). Within the surrounding region only LTBR and VAMP1 are over-expressed in this patient (Figure S6). For CD40, we identified one HMCL and two patient samples with spiked expression. By aCGH the HMCL XG-2 contains a discrete two-copy gain of the CD40 locus, which was confirmed by FISH and shown to be associated with the Ig-lambda locus (Figures S7). In the two patients with spiked CD40 expression, aCGH data were not available and cIg-FISH did not detect a copy number variability or rearrangement of CD40.

We also detected the spiked expression of TNFRSF13B/TACI in the HMCL KMS-26 and NFKB1 in OCI-MY5 and KMM-1. In both KMS-26 and OCI-MY5 the over-expression of TACI and NFKB1 were mirrored by the presence of homologous staining regions containing the respective genes (Figure 2). In KMM-1, which had a 19-fold increase in the expression of NFKB1, a detectable alteration by aCGH was not identified. However, by FISH an expression cloud was associated with a single NFKB1 allele that disappeared following RNase treatment (Figure S8), suggesting that an unknown mechanism is driving the over-expression of NFKB1 from a single allele in this HMCL.

Finally, we were able to correlate the presence of these diverse mutations with increased NF-kB transcriptional activity (Figure 2). The level of NF-kB mediated transcriptional activity was determined based on a gene expression index developed from the HMCLs. We used an ANOVA test using the error estimate from a cross-gene error model with multiple testing correction to identify 6 probesets (p<0.05) differentially expressed between the HMCLs with identified mutations compared to those without identified NFKB pathway mutations. We chose to remove two probe sets corresponding to HLA class II genes as their expression in patient samples may also derive from contaminating normal monocyte/macrophages and B-cells. The mean expression level of the remaining 4 probesets corresponding to CD74, IL2RG, and TNFAIP3 (2x) was used to represent the NFKB index. NF-kB transcriptional activity appeared to be high in all normal bone marrow plasma cells, likely due to stimulation from the surrounding microenvironment, but was variable in MM patients, and HMCL. Clearly, the presence of mutations in the NF-kB pathways was associated with a higher level of NF-kB transcription activity in the HMCLs. This correlation was also observed in our MM patients, albeit to a lesser degree, but the correlation is also evident in other datasets (Figure S4). The reduced correlation between the NF-kB index and mutations seen in the patient dataset may in part be due to the limitations inherent in the analysis of clinical material and to other sources of NF-kB activation originating from the bone marrow. However, in both datasets the association of these mutations with NF-kB transcriptional activity is not likely to result from chance alone as a hypergeometric distrubution test, a statistical test commonly used in enrichment analysis, returned p-values of 0.009 and 0.04 respectively.

FISH validation of genetic abnormalities within the NF-kB signaling pathways

To determine the frequency of TRAF3 deletions we blindly screened a large cohort of MM patients by cIg-FISH, including 60 patients initially screened by aCGH and 98 additional patients. We identified 6 (3.8%) patients with bi-allelic TRAF3 deletions and 19 (12.0%) patients with a single TRAF3 allele (Table 1). There was almost complete concordance between the FISH and aCGH results with the exception of two patients in whom the presence of the bi-allelic deletion was heterogeneous by FISH and below the detection limit of aCGH. However, in both these cases, aCGH demonstrated a clear loss of chromosome 14, which suggests that the bi-allelic deletions seen by FISH occurred after the q-arms of chromosome 14 were lost.

To systematically screen the limited patient material for deletions of TRAF2, cIAP1/2 and CYLD by cIg-FISH we selected patients for screening based on either 1N copy number of these genes, as determined by aCGH, or a median normalized expression below 0.4. This process identified three additional patients with cIAP1/2 deletions and one additional patient with a CYLD bi-allelic deletion (Table 1). No additional cases were identified with TRAF2 deletions. In summary, we identified bi-allelic deletions of TRAF2, TRAF3, CYLD or cIAP1/2 in 14/167 patients, and of TRAF3, CYLD or cIAP1/2 in 10/67 in the Carrasco dataset. We believe that the cumulative incidence of these abnormalities is under-estimated as only TRAF3 was studied in the majority of patients due to the limited availability of patient material.

Sequencing identifies mutations of TRAF3 and CYLD

To determine the overall rate of inactivation of the identified target genes we performed bi-directional sequencing of the coding regions of these genes. The individual coding exons and flanking mRNA splicing sites of TRAF3 were sequenced in 62 patients and in the entire panel of 46 HMCLs. Through this process, we found a third HMCL, KMS-11, with a bi-allelic deletion of TRAF3 originating from two independent deletions overlapping at exon 7 (data not shown). In the remaining HMCLs we found four independent mutations including three frameshifts and one in-frame deletion, which are predicted to inactivate TRAF3 (Figure 3, Tables 2 & S2). All HMCL, except the three HMCLs with bi-allelic deletions, expressed detectable full-length TRAF3 mRNA transcripts and no discernable transcript length variations, suggestive of unknown alternative splicing events, were observed.

We sequenced 49 randomly selected patient samples for TRAF3 and identified 3 different mutations (one nonsense and two missense). Subsequently, TRAF3 sequencing was extended to an additional 13 patients in whom LOH of the TRAF3 locus had been characterized by aCGH or cIg-FISH, and DNA or RNA was available for sequencing. This process identified 7 additional abnormalities within the coding region of TRAF3, including one nonsense, one frameshift, and four missense mutations along with one large deletion (Table 2 & S2). All of the missense mutations occur at amino acids conserved throughout vertebrate evolution (Figure S9), suggesting their functional relevance. Surprisingly, one of the patients with a subpopulation of cells with a bi-allelic deletion of TRAF3, MCR-0319, also had a missense mutation, which inactivates the RING domain of TRAF3 (Figures 3 & S9). A domain recently shown to be essential for the negative regulation of NIK by TRAF3 (He et al., 2006a). Altogether, TRAF3 is inactivated in 17% of HMCL and approximately 12.3% of MM patients by either bi-allelic deletion or by LOH associated with an inactivating mutation.

We sequenced the coding exons of CYLD in 11 patients with either low levels of CYLD expression or LOH and identified a single nonsense mutation, S265X, in one patient sample with LOH (Table 2 & S2). No mutations were identified in HMCLs with LOH of CYLD (Table 2). In the HMCL JK-6L, we identified a frameshift mutation in NFKB2, which was investigated based on the presence of an abnormal immunoblot band. In the only patient with LOH of the cIAP1/2 locus we did not detect an abnormality within the coding region of cIAP1 or cIAP2.

In summary, our comprehensive genetic screen identified a promiscuous array of genetic alterations that are predicted to result in constitutive activation of NF-kB signaling. When all of the bi-allelic deletion events, mutations, and gene rearrangements of TRAF2, TRAF3, cIAP1/2, CYLD, NIK, LTBR, CD40, TACI, NFKB2 and NFKB1 are combined, at least 28/167 (17%) MM patients and 19/46 (41.3%) HMCL have an abnormality that leads to the dysregulation of NF-kB signaling.

Genetic Alterations of TRAF3, BIRC2, BIRC3, CD40, NIK, TACI, LTBR correlate with the processing of NFKB2

To determine if the genetic aberrations we identified were associated with constitutive processing of NFKB2, we screened the panel of HMCLs for the processing of p100 to p52 (Figure 4 & Table S3). We identified constitutive NFKB2 processing in HMCLs with abnormalities of TRAF3, cIAP1/2, NIK, CD40, LTBR, and TACI. As expected, the HMCLs that over-express NIK or have TRAF3 inactivating abnormalities had constitutive NFKB2 processing while HMCL over-expressing NFKB1, OCI-MY5 and KMM-1, did not have detectable NFKB2 processing. The constitutive processing of NFKB2 from p100 into p52 correlates with both a high NFKB index and the detection of the normally cytoplasmic NFKB2 in the nucleus (Figure 4)(Table S3). In addition, the HMCLs with inactivation of TRAF3 and rearrangements of NIK were associated with detectable levels of NIK protein. Surprisingly, we also found high levels of NIK and constitutive processing of NFKB2 in HMCLs with cIAP1/2 bi-allelic deletions (Figure 4). The constitutive processing of NFKB2 seen in these HMCLs must be linked to the absence of cIAP1 and/or cIAP2 as the isogenic HMCL pair (KMS-28BM/PE), which were independently derived from the bone marrow or the pleural effusion of the same patient and share virtually identical genomes except for their difference in the presence (KMS-28BM) or absence (KMS-28PE) of cIAP1/2 also process NFKB2 differently. In this pair, the absence of cIAP1/2 is associated with increased levels of NIK protein and constitutive NFKB2 processing (Figure 4). As all of our bi-allelic deletions encompassed both cIAP1 and cIAP2 and we did not identify a mutation in either gene, we cannot determine if the loss of one or both is associated with NFKB2 processing.

Surprisingly, 2 of 3 HMCLs, L363 and EJM, with NIK abnormalities did not have detectable TRAF3 protein. However, the third cell line, JJN-3, which has a NIK rearrangement deleting the TRAF3 interaction domain (Pers. Comm. Michael Kuehl), had high levels of TRAF3 protein. Although it is known that TRAF3 regulates NIK protein stability, these two observations suggest that a direct NIK-TRAF3 interaction may also regulate TRAF3 protein levels. Furthermore, the high levels of TRAF3 in the cell lines lacking cIAP1/2 suggest this may occur in a cIAP1/2 dependent manner.

In conclusion, the majority of the genetic abnormalities we identified in our cohort of MM patients and HMCLs have highlighted a multi-faceted mechanism that constitutively activates the non-canonical NF-kB pathway. The selection of so many mutations in a single pathway highlights its importance in myelomagenesis and identifies a potential tumor specific vulnerability.

Characterization of TRAF3 as a Tumor Suppressor Gene

Since TRAF3 inactivation represented the most common abnormality we identified in the non-canonical NF-kB pathway in MM, we investigated its potential role as a tumor suppressor. We constructed an adenovirus capable of high efficiency infection, which expresses wild-type TRAF3 and used it to reintroduce TRAF3^wt into the HMCLs with TRAF3 abnormalities. Based on the well-documented role of TRAF3 as a negative regulator of NFKB2 processing we assayed its effects on p100 processing. In all of the HMCLs with TRAF3 abnormalities the reintroduction of TRAF3^wt was associated with an inhibition of NFKB2 processing from p100 to p52 (Figure 5). In addition, over-expression of TRAF3 in the HMCL L363, which over-expresses NIK due to an IgL translocation, leads to a marked inhibition of NFKB2 processing.

The involvement of TRAF3 in the regulation of NIK stability and the subsequent p100 processing is well documented, however, a role in oncogenesis has never been reported. We hypothesized that the inactivation of TRAF3 would provide cells with a selective growth advantage. As we predicted, the re-introduction of TRAF3^wt caused a significant growth delay in 4/5 HMCL associated with G0/G1 cell cycle arrest and a significant increase in apoptosis (Figure 5). The control HMCLs, OCI-MY5 and OPM-1, that do not have constitutive processing of NFKB2 were not affected by TRAF3 over-expression.

Inactivation of TRAF3 is associated with poor response to dexamethasone and a good response to proteasome inhibitors

Proteasome inhibitors are a novel class of anti-cancer agents with remarkable clinical activity in MM (Richardson et al., 2003). Although they presumably have multiple mechanisms of action, the principally defined mechanism has been postulated to be the inhibition of NF-kB signaling. We hypothesized that MM with constitutive activation of the non-canonical pathway would be particularly sensitive to proteasome inhibition. Unfortunately we could not test this hypothesis in the HMCLs, which are uniformly sensitive to proteasome inhibitors and differ from patients in a number of ways, most notably proliferation. Instead we examined a GEP dataset of relapsed/refractory MM patients randomly assigned to treatment with either dexamethasone or the proteasome inhibitor bortezomib (Mulligan et al., 2006). As we could not screen these patient samples for mutations, we used an ROC analysis of our patient dataset to identify a median normalized expression cutoff of 0.6 as the best crossover point between sensitivity and specificity and used this value as a surrogate means of identifying patients who likely have TRAF3 abnormalities. In our dataset 17/25 patients with TRAF3 expression below 0.6 lack functional TRAF3. Using this cutoff, 39/193 (20%) patients likely have a TRAF3 abnormality, a rate similar to that observed in our cohort. In this clinical trial dataset 2/20 (10%) patients with low levels of TRAF3 responded to dexamethasone whereas 17/19 (89%) responded to bortezomib. This was accompanied by a prolongation of progression-free-survival (PFS) from 83 to 193 days (p<0.0001). For patients with levels of TRAF3 above 0.6, 15/50 (30%) responded to dexamethasone and 42/104 (40%) to bortezomib, with no difference in PFS (140 vs 145 days)(Figure S10). These results suggest that constitutive activation of the non-canonical NF-kB pathway by inactivation of TRAF3 is associated with dexamethasone resistance and proteasome inhibitor sensitivity. The resistance to dexamethasone was unexpected but may relate in part to the fact that one of the mechanisms of action of glucocorticoids targets only the canonical NFkB pathway. It has been reported that hormone-bound glucocorticoid receptor is recruited to promoters by binding relA-containing dimers, thereby disrupting relA’s interaction with certain co-activators(Hoffmann and Baltimore, 2006). One would anticipate that this mechanism would be of little consequence in patients with constitutive activation of the non-canonical NF-kB pathway. Using the top and bottom third of the NFKB index to discriminate patients shows that patients in the high index group have a better response rate to bortezomib, although no difference in PFS. These results suggest that bortezomib may be most effective in patients with constitutive non-canonical NF-kB activation (as seen in patients with inactivation of TRAF3).

Sample Preparation

Bone marrow aspirates were collected from 167 MM patients (28 SMM and 139 overt MM). All patients provided written informed consent approving the use of their samples under Institutional Review Board approval. The BM aspirates were treated with ACK lysis buffer to remove red blood cells and plasma cells were isolated using immuno-magnetic sorting on a Miltenyi AutoMacs or StemCell Robocept with anti-CD138 antibodies. We enumerated 100 nucleated cells for kappa and lambda staining to determined the purity of the CD138 sorts and found that the mean plasma cell purity (kappa & lambda) was 95.5% (Median 96.0, range 69–100%) while the mean clonal purity (clinical light chain) was 92.7% (Median 95.0, range 54–100). Isolated cells were suspended in TRIzol (Invitrogen) and stored at −80°C for long-term storage. Nucleic acids were isolated from TRIzol following the protocol supplied by the manufacturer. The RNA used for gene expression profiling was cleaned up using the Qiagen RNeasy kit, while the DNA used for aCGH was cleaned up by phenol-chloroform extractions after RNase and proteinase K treatments. The HMCLs were maintained as previously described(Bergsagel et al., 1996).

Microarray experiments

GEP of the patient samples was performed on U133A genechips following the manufacturer's suggested protocol, while the 46 HMCL samples were done on U133 Plus 2.0 (Affymetrix, Santa Clara, CA) as previously described (Carrasco et al, 2006). Median normalized gene expression profiles were analyzed using GeneSpring (Agilent).

High-resolution aCGH was performed on 62 patient and 46 HMCL samples with the Human Genome 44B microarray (Agilent Technologies) following the protocol suggested by the manufacturer. Briefly, 800 ng of test (MM CD138+ cells or HMCL) and normal female reference (Promega) were directly labeled with either Cy5 or Cy3 using the BioPrime Array CGH Genomic Labeling Module (Invitrogen) and purified using Vivaspin 500 spin columns (Sartorius). The hybridization reactions containing equal amounts of test and reference DNA were hybridized to the microarray at 65°C for 40 h in a rotation oven at 20 rpm. The slides were washed and then scanned with the Agilent G2505B DNA microarray scanner. The microarray images were analyzed using Feature Extraction software V8.1 (Agilent) and Log2 transformed ratio data was analyzed with GeneSpring GX V7.3.1 and CGH Analytics V3.4.27 (Agilent)

Sequencing

When possible individual coding exons were amplified from 10 ng of CD138 purified plasma cell DNA in 25 ul reactions using Platinum Taq DNA Polymerase (Invitrogen, Carlsbad CA) and M13 tagged exon specific primes (Table S4). Alternatively, overlapping segments of the entire open reading frame were amplified from RNA converted to cDNA using Superscript II and oligo dT17. Sequencing reactions were carried out using Big Dye V3.1 and capillary electrophoresis was performed on an ABI 3730 at the Mayo Clinic or TGEN sequencing core facilities. DNA sequences were analyzed using Sequencher V4.5.

FISH

The cIg-FISH method was performed as previously described(Fonseca et al., 2003a). Briefly, probes specific to each region of interest were selected using the UCSC genome browser and the specific positions were confirmed by manual mapping of the available end sequences to data available from HGB 36.2. The selected BAC clones were purchased from Invitrogen and the Fosmid clones were ordered from the BACPAC Resources Center. The specificity of each probe was confirmed by hybridization to normal metaphase preparations to confirm chromosome and band specificity, and gene specific PCR was used to confirm the presence of DNA sequence corresponding to the specific region of interest. A list of the various FISH probes is provided in Table S5 and a map of the different regions is provided in Figure S11.

Immunoblotting, Immunofluorescence, and Immunohistochemistry

Total cellular extracts were prepared from freshly isolated HMCLs using 10x Cell Lysis Buffer (Cell Signaling Technologies) supplemented with 1mM PMSF. Protein samples were resolved on 7.5% SDS-PAGE gels and transferred to PVDF membranes. The individual blots were blocked with 5% non-fat milk for 1 hr and stained overnight with specific antibodies; TRAF3 H-122 (Santa Cruz), NIK (Cell Signaling), cIAP1 (human) and cIAP2 (human)(R&D Systems), NFKB2 (rabbit anti-p100/p52 N-terminal polyclonal, kindly provided by Warner Greene, UCSF). Protein bands of interest were imaged with HRP-conjugated secondary antibodies.

The localization of NFKB2 in HMCL and selected patient core biopsies was determined with an NFKB2 monoclonal antibody (18D10, Cell Signaling Technology). Fresh cytospins of HMCLs were fixed in 4% paraformaldeyde, permeabilized with 0.1% Triton X-100, and stained with a 1:50 dilution. Confocal slices of 2 μm were collected on a Zeiss LSM510 microscope using a 63x objective (Zeiss, Plan-Apochromat 63x/1.4 oil). Sections from available core biopsies were stained with a 1:50 dilution.

Generation of helper-dependent adenoviral vectors

We created a helper-dependent adenoviral (HDAd) vector precursor plasmid system capable of the simultaneous expression of EGFP and a gene-of-interest from two different promoters near the left and right ITRs, respectively. The original adenoviral plasmids, which have been described previously(Shi et al., 2006), were kindly provided by Mary Hitt (University of Alberta). First, we created the HDAd precursor plasmid, pSC27B, by replacing the beta-galactosidase expression cassette near the left ITR of pSC9B with a murine cytomegalovirus (mCMV) driven EGFP expression cassette. Second, we created the pSC11MCPA shuttle plasmid by introducing a mCMV-MCS-SV40 poly-A cassette in between the I-SceI and I-CeuI restriction sites of pSC11. Thus, allowing a gene-of-interest expression cassette from pSC11MCPA to be transferred to a region near the right ITR of pSC27B.

The full-length TRAF3 cDNA clone (Image Consortium clone 30915387) was purchased from Open Biosystems. The ORF was excised from the pCR4.0-TOPO vector with EcoRI and cloned into the MCS of pSC11MCPA. The expression cassette was excised from pSC11MCPA-TRAF3 with I-CeuI and I-SceI and transferred to pSC27B to create pSC27B-TRAF3. Helper-dependent adenoviral particles were prepared by repeated cycles of amplification in 293Cre4 cells (kindly provided by Frank Graham, McMaster University) with the adeno helper virsus AdRP2050 (Robin Parks, Ottawa Health Research Institute), which contains a capsid modified by insertion of RGD into the H1 loop of the fiber. Virus particles were purified with a 1 hr step gradient and overnight continuous CsCl gradients as described previously (Parks et al., 1996). Virus particles were quantified as described previously (Ng et al., 2002). Healthy HMCLs were infected with 150 viral particles per cell in 60mm Petri plates with 2.5×10⁶ cells in 5 ml of antibiotic free media. The percentage of infected cells was monitored by FACS analysis of the co-expressed GFP marker.

Functional Analysis of TRAF3^wt reintroduction

Infected HMCL were counted 24 hours post-infection and seeded at specific densities in triplicate in 96 well plates, for MTT assay, and 6 well dishes; for protein assays, apoptosis, and cell cycle analysis. For protein assays TRAF3^wt and control infected cells were isolated 48 hours post infection. For cell cycle analysis cells were stained with propidium iodide and the FACS results were modeled with ModFit LT and chi-square value below 5. The number of apoptotic cells was determined by Annexin-V-Alexa 647 (Invitrogen) staining.

Electronic availability of the Data

A complete list of the patient samples and various tests are provided in Supplementary Table S6. The publicly available datasets used in this paper are available in gene expression omnibus (GEO) or institutional websites as follows, Carrasco et al. GEP (GSE4452) and aCGH (http://genomic.dfci.harvard.edu/array_cgh_db.htm). The GEP data from our patient dataset is available in the GEO database under the accession number GSE6477. The complete dataset is freely available from the Multiple Myeloma Research Consortium Genomics Portal (http://www.broad.mit.edu/mmgp).