Introduction

Diffuse large B-cell lymphomas (DLBCLs) represent the most common subtype of non-Hodgkin lymphoma (NHL).1 Gene expression profiling classifies DLBCL into at least two molecular groups, including the germinal center B-cell (GCB) and activated B-cell (ABC) subtypes.2 GCB DLBCLs are characterized by the overexpression of genes active in normal germinal center B cells, such as BCL6, CD10 and CD38; while ABC DLBCLs overexpress genes normally induced during B-cell activation, such as MUM1, c-FLIP and BCL2.2

BCL2, discovered because of its involvement in t(14;18) in follicular lymphoma (FL),3 has a central role in the inhibition of apoptosis.4 BCL2 is normally transiently expressed during B-cell maturation. The t(14;18) translocation causes constitutive overexpression of BCL2 by juxtaposing it to immunoglobulin heavy chain gene enhancer elements. This translocation is found in ∼20% of DLBCL,5 most often in the GCB subtype of DLBCL.6 Other mechanisms of BCL2 deregulation, more often observed in ABC DLBCLs, include amplification of the BCL2 gene or its transcriptional upregulation through constitutive activation of the nuclear factor-κB pathway.7 Saito et al.8 reported that the BCL2 promoter can also be aberrantly hypermutated in DLBCL, which may prevent its inhibition by MIZ1 and BCL6.8 They found a high number of mutations in BCL2 and suggested that the mutations described were the result of somatic hypermutation (SHM). Mutations in BCL2 have been shown to cause false-negative protein expression results in immunohistochemistry assays in FL;9, 10 however, no large study has so far investigated whether this occurs in DLBCL.

BCL2 protein overexpression in DLBCL has been associated with poor prognosis.11 The addition of rituximab to chemotherapy may overcome the impact of BCL2 expression on prognosis;12 however, BCL2 expression remains relevant when analysis is restricted to specific molecular subtypes of DLBCL.7, 13, 14 Indeed, drugs that target BCL2 are being investigated in clinical trials, thus BCL2 mutations may be clinically important in the future.

The BCL2 family of proteins comprises members with both pro- and anti-apoptotic roles. All have at least one of the four structurally conserved motifs known as BCL2-homology (BH) domains.15 BCL2, BCL-XL and other anti-apoptotic members contain all four BH domains, while pro-apoptotic members (for example, BAX and BAK) either contain BH1, BH2 and BH3, or BH3 alone (the so-called ‘BH3-only’ class). The BH3 domain of pro-apoptotic members inserts into a hydrophobic groove domain formed by the BH1, BH2 and BH3 domains of anti-apoptotic members, a key interaction for the negative regulatory role of BCL2 family members.16

Using the transcriptome sequencing method RNA-seq17 in 89 DLBCL samples, we recently identified genomic changes that drive the pathogenesis of adult DLBCL.18, 19 The most commonly mutated gene, mutated in 33/89 transcriptomes (37%) was BCL2.19

The aims of the current study were to determine the frequency and distribution of BCL2 mutations in DLBCL in a large number of cases, assess the relationship between gene mutation and BCL2 protein expression or BCL2 translocation, and determine the impact of mutations on immunohistochemical detection of protein expression and clinical outcome. We also tested whether BCL2 mutations are specific to DLBCL by sequencing the BCL2 gene in other lymphoma subtypes with and without BCL2 translocations. Re-sequencing BCL2 in 298 DLBCL tumors revealed that it is highly mutated in this NHL subtype, showing that alteration of BCL2 in DLBCL can occur not only by the established mechanisms of translocation or amplification, but also, frequently, through specific point mutations.

Materials and methods

Samples and patients

In all, 512 samples were used: 429 primary tumor biopsies, 24 DLBCL cell lines, 51 germline DNA samples and 8 tonsil CD77+ centroblast samples. Primary tumor samples consisted of 298 DLBCL (Table 1), 25 MCLs (mantle cell lymphomas), 25 peripheral T-cell lymphomas (PTCLs), 30 SLLs (small lymphocytic lymphomas), 26 FLs and 25 PMBCLs (primary mediastinal B-cell lymphomas). All samples were taken before treatment and diagnosed according to the World Health Organization 2008 criteria by expert hematopathologists.1 Cell lines and their respective sources are listed in Supplementary File 1. In a subset of 51 of the above individuals (25 patients with DLBCL and 26 patients with FL), normal peripheral blood lymphocytes were used as a source of germline DNA. In addition, CD77+ centroblasts purified from pediatric and adult tonsils using Miltenyi magnetic bead separation according to the manufacturer's protocol (Miltenyi, Auburn, CA, USA) were used as a source of normal germinal center B cells. Patients with DLBCL were treated with curative intent with cyclophosphamide, doxorubicin, vincristine and prednisone (CHOP) with (n=212) or without (n=86) rituximab (R) (Table 1). Ethical approval for this study, conducted in accordance with the Declaration of Helsinki, was granted by the University of British Columbia/British Columbia Cancer Agency Clinical Research Ethics Board.

Table 1 Clinical characteristics of samples subjected to Sanger sequencing, categorized according to the presence of coding mutations in BCL2
Full size table

Immunohistochemistry

Standard immunohistochemical methods were used for all antibodies. BCL2 protein expression was assessed in 262 DLBCL samples. Whole sections of FFPET (fixed paraffin-embedded tissue) were stained with antibody clone 124 (Dako, Burlington, ON, Canada), which targets amino acids 41–54. In all, 135 (52%) of these samples were also included in tissue microarray that was stained with antibody clone 124 and clone E17 (Epitomics, Burlingame, CA, USA), which targets amino acids 50–90 (96/135 samples were previously reported20, 21). Samples with 50% or more malignant cells expressing BCL2 were defined as BCL2 positive.

In all, 69 DLBCL cases were stained with CD10, BCL6 and MUM1 to determine cell of origin subtype by Hans criteria.22 Gene expression data for cell of origin molecular subtyping were available for an additional 132 DLBCL cases, totaling 201/262 cases with cell of origin information (Table 1). Immunohistochemical stains were independently evaluated by three hematopathologists and samples with discrepant results were reconciled by consensus review using a mutliheaded microscope.

Cytogenetic analysis

In 175 cases, the presence or absence of a BCL2 translocation could be confirmed (Table 1) by FISH (fluorescence in situ hybridization) (n=164), using a commercial Vysis dual color break-apart probe (Abbott Molecular, Abbott Park, IL, USA) applied to FFPET, or by G-banded karyotype (n=11).23 For samples in which non-synonymous BCL2 mutations were detected, FFPET were retrieved for FISH analysis. Samples containing >5% of cells displaying break-apart signals were considered positive for the presence of a translocation. The break-apart probe identifies BCL2 translocations, and while t(14;18) is most common, translocations with (a) IGK (κ light chain, 2p11.2); (b) IGL (λ light chain, 22q11.2) and (c) complex variant translocations involving chromosomes 14, 18 and one or two other chromosomes24 would also be detected using this assay.

Transcriptome sequence data

RNA-seq data included 31 DLBCL libraries (each RNA-seq library is constructed from a single tumor sample) described by Morin et al.18 and 58 additional DLBCL libraries.19 Briefly, paired end reads, obtained by sequencing both ends of each DNA fragment of an RNA-seq library, were aligned to the human reference genome (hg18) and exon junction sequences (to allow correct alignment of reads spanning exons) using the Burrows-Wheeler Aligner.25 Single nucleotide variants were identified with SNVMix software (http://compbio.bccrc.ca/software/snvmix/).26 Known polymorphisms, including those found in dbSNP (build 130) and any variants in published genomes, were removed, resulting in a list of candidate mutations in the genome, including in BCL2.

Sanger sequencing

For all samples described in ‘samples and patients,’ genomic DNA was extracted using the QIAGEN ALL PREP kit (Qiagen, Hilden, Germany). DNA was quantified using PicoGreen (Invitrogen, Burlington, ON, Canada) in a Victor2 fluorescence plate reader (Perkin-Elmer, Waltham, MA, USA). Four BCL2 amplicons were sequenced, with primers designed using Primer3(ref. 27) to cover all exons and 1.4 kb upstream of the BCL2 transcriptional start site. Supplementary File 2 lists each amplicon's primers and PCR conditions. –21M13F (TGTAAAACGACGGCCAGT) and M13R (CAGGAAACAGCTATGAC) extensions were added to 5′ ends of forward and reverse PCR primers, respectively, for uniform sequencing conditions. PCR products were purified using AmPure magnetic beads (Agencourt Bioscience, Beverly, MA, USA) and eluted in 30 μl Tris EDTA (10 mM Tris, 0.1 mM EDTA pH 8.0) according to the manufacturer's instructions. In all, 2 μl of purified product was cycle sequenced using Big Dye Terminator Mix V.3 at 1/24 chemistry in total volume of 4 μl (Applied Biosystems, Foster City, CA, USA). Both forward and reverse directions were sequenced. Mutation Surveyor v3.24 (Soft Genetics, State College, PA, USA) was used to assemble against build hg18 and examine the sequences.

In-silico functional assessment of BCL2 mutations

PolyPhen,28 a tool for in-silico classification of amino-acid changes as ‘benign’, ‘possibly damaging’ or ‘probably damaging’, was used to predict possible deleterious effects of non-synonymous mutations on BCL2 protein. ‘Damaging’ variants affect annotated active sites or post-translational modification sites, or have predicted effects on the three-dimensional structure of the protein.

Statistical analysis

Progression-free survival (PFS) was determined from the date of diagnosis to the date of relapse, progression or death of any cause. Overall survival (OS) was determined from the date of diagnosis to the date of death of any cause. The median follow-up in patients alive at last follow-up was for a median 5.7 years (0.1–15.9 years). OS and PFS were calculated using the Kaplan–Meier method and the log rank test was used to compare survival curves between groups.20 The χ2 test and Fisher's exact test were used to test for association between variables; P-values of <0.05 were considered significant. Unless otherwise specified, P-values reported are calculated from Fisher's exact test. These statistical analyses were performed using SPSS software, version 14 (SPSS Inc., Chicago, IL, USA).

Results

Frequent somatic BCL2 mutations in DLBCL and FL

The median depth of coverage of BCL2 for all RNA-seq libraries was 58-fold. The BCL2 gene was identified as the most mutated expressed gene in DLBCL transcriptomes19 (Figure 1), with 33 of 89 (37%) DLBCL samples containing at least one candidate mutation in the coding region, 26 of which had at least one non-synonymous mutation. On average, 5.53 mutations affected the BCL2 coding region in the 26 mutated DLBCL transcriptomes. BCL2 was mutated twice as often as the next most commonly mutated gene, PIM1 (Table 2), and 5.6 times more than EZH2. There were far more mutations in BCL2 than in other genes known to be targets of SHM, such as PIM1 or BCL6,29 which were only mutated in 25/89 and 8/89 samples, respectively (Table 2). Of known SHM targets, only BCL2 had significantly more mutations in GCB samples (P=0.004; Table 2). Of the 10 other BCL2 family genes consistently expressed (with expression defined as mean RPKM17⩾20), the average number of samples with a candidate mutation was 1; only two other genes in the family, MCL1 and BCL2L11, had candidate mutations in more than two samples.

Figure 1
figure 1

Mutation frequency in 28 commonly mutated genes in B-cell NHL. The bar height is the total number of non-synonymous candidate mutations in each gene detected by RNA-seq in 89 cases. These 28 genes include 26 described by Morin et al. as significantly mutated, and two genes (CARD11 and TNFAIP3) also known to be targets of mutation in DLBCL.

Full size image
Table 2 Candidate mutations in known targets of somatic hypermutation identified by RNA-seq in DLBCL cases
Full size table

The high frequency of BCL2 mutations was then determined in a larger set of NHL tumors (of DLBCL, MCL, SLL, FL, PMBCL and PTCL subtypes) by Sanger sequencing. All three exons and 1.4 kb upstream of the BCL2 transcriptional start site were successfully bi-directionally sequenced in 97% of samples. After exclusion of known polymorphic variants (rs1801018, rs1473418 and rs2279115), 594 heterozygous mutations were detected in 104 of the 298 (35%) primary DLBCL samples (Figure 2 and Supplementary File 3). This high mutation rate is only sixfold lower than that of the rearranged IGH gene (9 × 10−4 mutations/bp for BCL2, compared with 5.8 × 10−3 mutations/bp for IGH30). Of these, 127/594 were single nucleotide substitutions in the coding region of BCL2, of which 60/594 (10%) were non-synonymous, predicting amino-acid substitutions or truncation of the BCL2 protein in 38 samples. Of note, some samples contained multiple non-synonymous mutations (n=4 in 2 samples, n=3 in 3 samples, n=2 in 9 samples, n=1 in the remaining 24 samples). In addition, 115 positions had mutations in more than one case (296/594 mutations were in these recurrent positions). The most frequently observed non-synonymous mutations occurred at amino acids 59 and 60 (9/38 samples), in the flexible loop domain (FLD). Heterozygous insertions/deletions (indels) were detected in 9/104 samples; none was found in the coding region.

Figure 2
figure 2

Mutations found in NHL tumors by Sanger re-sequencing. The number of mutations at each position is shown on the y axes; distance along the BCL2 gene is indicated on the x axis. A model of the BCL2 gene is shown beneath the x axis of the top left figure, 5′–3′ from left to right.

Full size image

Mutations in BCL2 were also frequently detected in FL, but not SLL, PTCL, PMBCL or MCL (Figure 2). Sequencing BCL2 in 26 FL samples revealed 139 mutations, including 5 indels in introns. In contrast, only 11 heterozygous mutations were detected in the other 105 samples comprising SLL, PTCL, PMBCL and MCL. Sequencing of matched constitutional DNA (from 26 FL and 25 DLBCL cases) and 8 normal centroblasts revealed, respectively, one heterozygous synonymous variant and four additional variants (two being the common SNPs (single nucleotide polymorphisms) rs2279115 and rs1801018, while the other two were synonymous variants), implying that >99% of sequence variations observed in GCB-type DLBCL and FL samples are somatic mutations specific to the tumor. Matched tumor DNA for the 25 DLBCL cases had an average of 5.7 mutations per sample (range: 0–28), while matched tumor DNA for 26 the FL cases had 5.4 mutations per sample on average (range: 0–19), with only 4/26 FL tumor DNA having no mutations.

In all, 23/24 DLBCL cell lines also had mutations (Supplementary File 4), with an average of 11 BCL2 mutations per cell line (range: 1–79). Compared with tumor samples, cell lines had many homozygous mutations, with 42% of mutations homozygous.

BCL2 mutations are associated with GCB phenotype and presence of BCL2 translocation

BCL2 mutations were significantly enriched in GCB DLBCL and highly associated with the presence of BCL2 translocation (see Figure 2 and Table 1). Cell of origin molecular subtype information was available for 201 DLBCL samples (95 GCB and 106 ‘non-GCB’, which included 44 cases of ABC and 22 cases unclassifiable as defined by gene expression profiling and 40 cases of non-GCB defined by the Hans immunohistochemical algorithm22). Mutations were associated with GCB vs non-GCB tumors (P<0.00001) (Table 1). In addition, only 7/44 ABC samples (16%) had any mutation in BCL2 (synonymous or non-synonymous) (range: 0–6), while 61/95 GCB samples (64%) had at least one mutation (range: 0–51, P=7.34 × 10−8). Two of the seven ABC samples with BCL2 mutations also had a BCL2 translocation by FISH. Translocations of BCL2 were detected in 25/26 FL (96%) and in 61/172 DLBCL (35%) (Table 1), and were significantly associated with the presence of BCL2 mutations (P<0.00001).

If these mutations occurred as a result of SHM, one would expect a high transition/transversion ratio from activation-induced deaminase activity, which removes amino groups from cytosines, creating C>T changes.31 If the mutations were randomly generated, one would expect a 1:2 transition/transversion ratio. Indeed, in DLBCL samples, the transition/transversion ratio was inverted (343/241, 1.42:1), thus demonstrating a clear enrichment for transitions (P=4.07 × 10−18) supporting previous reports8 that BCL2 mutations likely occurred as a result of SHM. The transition/transversion ratio was similar for data obtained by RNA-seq and Sanger methodologies. Consistent with SHM, we observed ‘clustering’ of mutations (see Figure 2) at promoter sites. In our large set of cases, we observed that this clustering of mutations is most pronounced near P1 (the more 5′ promoter) in GCB tumors. Clustering is apparent near both promoters in FL and, interestingly, the few mutations seen in MCL, PTCL or SLL are close to P2. The large number of mutations close to the promoters, with a tail-off in the distribution with increasing distance from the promoter, is characteristic of SHM. Transitions were most enriched in the GCB samples (ratio of 1.48:1) compared with the ABC cases (ratio of 0.63:1), where there was no evidence of enrichment for transitions.

The high transition/transversion ratio in BCL2 translocation-positive cases (P=2.50 × 10−13) compared with BCL2 translocation-negative cases (P=0.07) supports the previous report of the BCL2 translocation in placing BCL2 at high risk of acquiring mutations through SHM.32 These data strongly suggest that presence of BCL2 translocation, rather than deregulation of BCL2 through other mechanisms such as transcriptional upregulation seen in ABC subtypes of DLBCL or deregulation of miR 15–16 in CLL, has a pivotal role in the acquisition of BCL2 mutations in GCB DLBCL and FL. However, 17% of DLBCL cases without a detectable BCL2 translocation contained mutations of BCL2, indicating that other mechanisms may also induce mutations in BCL2. These other mechanisms appear to be tied to BCL2 expression level: in the RNA-seq data BCL2 mRNA levels where higher in BCL2 mutated cases than in cases without BCL2 mutations (Supplementary File 5), when comparing only cases with no evidence for BCL2 translocation (t-test P=0.016). Furthermore, BCL2 protein status was available in eight cases with BCL2 mutations and no BCL2 translocation. Although the numbers in this subset analysis are too small to be statistically significant, 6/8 cases were positive for BCL2 protein, indicating BCL2 is indeed expressed in samples without translocation. The lack of BCL2 mutations in germline DNA and normal centroblasts is consistent with previous reports33 that, unlike BCL6, BCL2 is not a target of SHM in normal tissue.

Non-synonymous mutations are enriched in the FLD, between the BH1 and BH3 domains and at the amino terminus

Non-synonymous mutations in BCL2 were significantly enriched in three areas corresponding to the amino terminus, the FLD and amino acids 108–135 located between the BH1 and BH3 domains of the protein, compared with all other domains (Figure 3). All mutations, their predicted amino-acid changes and, where applicable, their dbSNP identifiers, are provided in Supplementary File 3. Only 9/60 non-synonymous mutations were detected in the BH domains, which are important for inhibiting apoptosis. In contrast, 18 non-synonymous mutations were found in sites that would affect the BCL2's putative p53-binding domain34 (Figure 3), including an annotated caspase cleavage site at D34. D34 mutants have reduced ability to induce apoptosis,35, 36 including p53-mediated apoptosis.34 8/51 non-redundant, non-synonymous mutations were predicted to be ‘probably damaging’ by PolyPhen, while an additional 15/51 were predicted to be ‘possibly damaging’ (Figure 3). For instance, non-synonymous mutations at R146 may impair the interaction between BCL2 and BAX.37 One variant introduced an early stop codon, which would result in a form of BCL2 with no transmembrane domain.

Figure 3
figure 3

Non-synonymous mutations found in the BCL2 protein. Functional domains of the protein are indicated in different colors and are named on the figure; mutations in a given domain are shown in the same color. Mutations predicted by PolyPhen to be damaging to BCL2 function are indicated as triangles. TM, transmembrane domain.

Full size image

BCL2 mutations are not a common cause of pseudo-negative BCL2 protein expression in DLBCL

The large sample size in this study allowed us to determine that ‘pseudo-negative’ BCL2 protein expression due to BCL2 mutations is not frequent in de-novo DLBCL. Both synonymous and non-synonymous BCL2 mutations were associated with BCL2 protein expression (P=0.0007). In 135 cases, both monoclonal antibodies 124 and E17 could be compared. Only 13/135 cases (10%) were discordant for staining with clones 124 and E17 (Table 1). Of nine cases negative for clone 124 and positive for E17, none had non-synonymous mutations that could interfere with the binding site for clone 124. Four cases were considered negative for clone E17 and positive for clone 124 but no non-synonymous mutations were detected in these cases. Variability in tumor content (percent cells positive) may account for these discrepancies. Thus, although ‘pseudo-negative’ BCL2 protein expression from BCL2 mutations has been reported in FL before and after histological transformation to DLBCL, this is a rare event in de-novo DLBCL.9

BCL2 mutations do not confer a more favorable OS

Patients with BCL2 mutant DLBCL do not have better clinical outcome compared to patients with BCL2 wild-type DLBCL, suggesting that BCL2 mutations do not result in loss of anti-apoptotic function or improved chemosensitivity. In this study, BCL2 protein expression was associated with inferior survival in CHOP-treated (P=0.003) but not R-CHOP-treated patients (P=0.807), consistent with previous publications.12 We focused on prognostic effects of BCL2 mutations and conducted comparisons when sufficient numbers of samples with and without each variable were available. The presence of a BCL2 translocation or non-synonymous mutation was not significantly associated with OS or PFS in CHOP or R-CHOP-treated patients (all P-values>0.05). We also tested within the GCB subtype and within cases that had a BCL2 translocation, and within cases with BCL2 non-synonymous mutations located within the FLD (in the presence or absence of a BCL2 translocation). No impact was found on OS or PFS. In a Cox regression model that included International Prognostic Index, use of rituximab, presence of BCL2 expression and presence of mutations within the FLD, the effect of BCL2 mutations was also not statistically significant. In some analyses, the number of cases included was too small to make firm conclusions about the prognostic significance of BCL2 mutations. However, overall, BCL2 mutations do not seem to have an impact on patient outcomes.

Given that novel BH3 mimetics, for example, ABT-737, are currently being evaluated in patients with lymphoid malignancies, it is important to determine if BCL2 mutations may be a potential cause of resistance to these agents. Of the 24 cell lines included in this study, we found non-synonymous mutations in the FLD of BCL2 in three cell lines that have been previously evaluated for responses to ABT-737.38 These included KARPAS 422 (with the mutations G47D and P59L), OCI-LY1 (with F49L mutation) and SU-DHL6 (with I48IF mutation). All three are very sensitive to ABT-737; thus mutations in the FLD may not interfere with the binding of BH3 mimetics. Further work would be necessary to show whether these or other mutations impair the binding of BH3 mimetics to BCL2.

Discussion

The high frequency and non-random pattern of BCL2 mutations that we and others8 have observed supports the notion that these are noteworthy events in DLBCL. BCL2 mutations have been previously described in DLBCL and FL, but their frequency, compared with other genes mutated in the DLBCL genome, had not been established. Furthermore, previous studies have combined cell line data in analysis with primary patient samples, making results hard to interpret.

With novel sequencing technologies, we demonstrated that the number of BCL2 mutations is at least threefold higher than that of the second most mutated gene, PIM1, and only sixfold lower than that published for rearranged IGH.30 We show that BCL2 is mutated at a higher rate than all known targets of SHM29 except IGH in GCB DLBCL. Our data support other reports that BCL2 mutations are associated with the presence of a BCL2 translocation and likely occur as a consequence of aberrant SHM,8, 32, 39, 40, 41 suggesting that the BCL2 locus may have properties that hinder DNA repair machinery when occupied by other proteins in a GCB-type specific manner. However, 6/39 (15%) of the cases with non-synonymous mutations in this study occurred in the absence of a BCL2 translocation, suggesting that rearrangement is not a requirement for BCL2 mutations.

The high transition/transversion ratio suggests that aberrant SMH is a predominant feature of the GCB-type DLBCL but, unlike previous work,8 we do not find evidence for SHM in ABC-type DLBCL or normal germinal center B cells. This could be due to the larger number of ABC DLBCL samples in our study. Other targets of aberrant (and normal) SHM have been described, including BCL6, PAX5, MYC and RHOH.29 These have all been suggested as uniquely hypermutated in DLBCL and not in other NHLs. Our transcriptome data revealed mutations in all of these genes except RHOH, but of these only BCL2 showed a significant enrichment for mutations in the GCB subtype (P=0.004; Table 2). There are three possible causes for the BCL2 mutations observed: SHM machinery could be targeting (a) the IGH locus, (b) the BCL2 locus—targeting of which our data does not support since we do not observe mutations in normal germinal center B cells and (c) IGH/BCL2 sequence as brought together by t(14;18). BCL2 may be the only gene preferentially mutated in the GCB subgroup because BCL2 translocations are known to be associated with GCB classification.6 After BCL2, PIM1 showed the next highest number of potentially functional mutations, with non-synonymous single nucleotide variants in 17 out of 89 DLBCL samples. Notably, the single tumor with the most mutations in PIM1 (4 non-synonymous, 11 total) was of the ABC subtype. It is possible that mutations reflect the outcome of a cascade of events in which cell type origin affects chromatin configuration and gene expression, which influences the likelihood that a specific gene is available for translocation and/or mutation.

Our sequence analysis of BCL2 in over 400 samples revealed a distribution of non-synonymous mutations (see Figure 3) that was not apparent in smaller sample sets,8, 32, 39, 40, 41 the largest of which sequenced ∼3 kb in 121 samples.8 We found very few non-synonymous mutations in the BH domains required for BCL2 to block pro-apoptotic family members, but did find many mutations in the FLD. The FLD can act as an auto-inhibitory domain reducing anti-apoptotic potential.42 It contains the caspase-mediated cleavage signal at D34 (that we found mutated in DLBCL patients), which renders BCL2 pro-apoptotic.35, 36 The FLD also has important roles in paclitaxel-induced apoptosis,43 cell-cycle progression,44, 45, 46 and interaction with Nur77, a cell death-related protein.37 The transmembrane domain of BCL2 is required for Nur77 to trigger cell death; interestingly, we found a truncating mutation in one sample, which would then lack this domain.

Almost half of the observed non-synonymous mutations were predicted by PolyPhen to affect protein function. Although 28/51 unique non-synonymous mutations were marked ‘benign’ by PolyPhen, this does not preclude a functional role. For example, non-synonymous mutations at residue A43 were scored as ‘benign’ by PolyPhen; but previous work47 has shown that changes at this position provide resistance to autoimmunity via reduction of T cells.

The most common mutations observed in this study (amino acids 59 and 60) are located in a regulatory binding site for p53.34 BCL-XL binds p53 through its own region homologous to the FLD of BCL2, as well as through the loop between a3 and a4.48 The helices a3 and a4 are located between BH3 and BH1 domains, and we observed many non-synonymous mutations between these domains of BCL2, with four cases containing non-synonymous mutations between the two helices (amino acids 117–123).

Phosphorylation of BCL2 affects several BCL2 functions, depending on cellular context, including survival,49 autophagy50 and cell-cycle control.51 It would therefore be interesting to further study the functional effect of non-phosphomimetic mutations at S87, such as S87R reported here.

Another novel aspect of this study is that it addresses the potential clinical ramifications of BCL2 mutations in DLBCL. Most clinical laboratories use BCL2 antibody clone 124 from Dako (Abbott) to test for BCL2 protein expression. Mutations in the FLD have been reported to result in ‘pseudo-negative’ or ‘false-negative’ BCL2 expression,10 where BCL2 could be detected with an alternative anti-BCL2 clone E17,9 suggesting that the E17 clone is more accurate for determining BCL2 protein status. However, previous studies were done using selected cases and with relatively small sample sizes. We find that ‘pseudo-negative’ BCL2 protein expression with clone 124 is rare in de-novo DLBCL. We therefore do not support recent arguments, suggesting that pathology laboratories change antibodies.9

As a group, BCL2 mutations do not appear to be associated with clinical outcome; however, a larger sample size would be required to confirm this. Should a larger cohort become available for such an analysis, sequencing the regions rich in mutations in the amino terminus of the protein is likely to provide sufficient data. We also found that three cell lines previously described as sensitive to ABT-737(ref. 38) harbor BCL2 mutations in the ABT-737-binding area, indicating that these mutations are not sufficient to affect sensitivity to BH3 mimetics. Interestingly, the best clinical responses to ABT-737 or ABT-263 were found in patients with CLL,52 a lymphoma subtype that does not harbor mutations of BCL2.

In summary, we report that BCL2 is the most commonly mutated gene in GCB-type DLBCL. It is difficult to predict the effect of mutations on BCL2 function due to its pleiotropism, with known roles in autophagy, ER Ca2+ storage,53 cell-cycle entry and apoptosis, in which it can act as protector or killer. BCL2 function also varies depending on which genes are co-expressed, and whether those genes are also mutated.37, 54

BCL2 is a cell fate master switch, and a large portion of its decision-making may rest on its large FLD. Systematic functional characterization of BCL2 mutants will be necessary to understand the effects on cell fate decisions, B-cell development, lymphomagenesis and potentially treatment response.