BCL2 Predicts Survival in Germinal Center B-cell–like Diffuse Large B-cell Lymphoma Treated with CHOP-like Therapy and Rituximab

See commentary by Dunleavy and Wilson, p. 7505

Diffuse large B-cell lymphoma (DLBCL) is the most common type of non–Hodgkin lymphoma, with several morphologic and clinicopathologic variants (1). Distinctive molecular and genetic abnormalities have been identified in DLBCL, and patients with this disease exhibit a wide range of clinical presentations and outcomes (2, 3). The International Prognostic Index (IPI) is a widely accepted tool used to predict the clinical outcome of the patients with DLBCL (4). However, new therapeutic regimens with better effectiveness can alter the significance of prognostic markers, and modifications of the IPI have recently been proposed to better predict the outcome in DLBCL (5).

Gene expression profiling (GEP) studies in cyclophosphamide-Adriamycin-vincristine-prednisone (CHOP) era have shown that patients with DLBCL derived from germinal center B cells (GCB-DLBCL) have a better survival than those patients with DLBCL derived from activated B cells (ABC-DLBCL; ref. 6). However, the addition of the anti-CD20 antibody, rituximab (R), has revolutionized the treatment of DLBCL, leading to a significant increase in survival (7), but survival advantage in patients with the GCB-DLBCL persists (3, 8, 9).

BCL2 functions as an antiapoptotic factor and is frequently deregulated in DLBCL (10). One well-characterized mechanism of BCL2 overexpression is the t(14;18)(q32;q31), which is largely restricted to GCB-DLBCL (10), whereas in ABC-DLBCL, the mechanism of BCL2 overexpression is associated with constitutive NF-κB activation (11), with or without 18q21 amplification (12). BCL2 has been extensively studied as a prognostic biomarker in DLBCL, albeit with controversial findings (13–19) which were thought to be due to the heterogeneity within DLBCL (12, 20). Our previous studies of patients with DLBCL treated with CHOP-like therapies have shown that BCL2 expression has prognostic significance in the ABC-DLBCL but not in GCB-DLBCL (12, 21).

Because the standard treatment of DLBCL now includes rituximab, several studies with conflicting observations concerning prognostic value of BCL2 expression in R-CHOP–treated patients have been reported in recent years with some showing (22, 23) and some not (24, 25) showing any prognostic significance. Therefore, we analyzed our cohort of GEP-defined DLBCL to determine whether BCL2 overexpression still has prognostic value in patients with DLBCL treated in the R-CHOP era.

Two-hundred and twenty-one cases of de novo DLBCL treated with rituximab and CHOP or CHOP-like therapies were obtained from the LLMPP consortium (3). The CHOP cohort (n = 181) was used for comparative analysis (3, 12) only. This study was approved by the Institutional Review Boards of the respective institutions, and all patients gave written informed consent.

Of the R-CHOP cohort, 169 cases were evaluated for BCL2 protein expression, using BCL2 antibody (Clone-124; Dako) as reported previously (12, 26). Approximately half of the cases (tissue microarrays) had been scored previously by 2 pathologists (W.C. Chan and D.D. Weisenburger; ref. 27), and these cases were rescored (by P.N. Meyer) with no significant disagreement. The rest of the cases were then evaluated by P.N. Meyer, and his scores were used for this study. The tumor cell percentage was recorded in 10% increments, and 50% or more staining was considered positive. The optimal cutoff point for BCL2 expression (positive vs. negative in survival analysis) was determined by survival tree methods using P values from adjusted log-rank statistics. The adjusted log-rank statistics takes into account that a large number of cutoff points may be tested when determining the optimal cutoff point (28). In this approach, optimal cutoff point for biomarker is selected as the one with the maximum adjusted log-rank statistics to predict patient overall survival (OS) in all data, and then this cutoff point is applied for subgroups analysis. The survival trees were created with the “party” package: A Laboratory for Recursive Partitioning in rituximab (29, 30).

GeneChip HG-U133 plus2.0 array (Affymetrix Inc.) was previously used for GEP for determining the cell of origin in both R-CHOP and CHOP cohort (3). BCL2 mRNA expression was measured from mean intensity of 3 probesets (232210_at, 244035_at, 232614_at) out of the 8 probesets present on HG-U133 plus2.0 array. These probesets were chosen on the basis of their consistence in measurement, the signal intensity, and their ability to measure BCL2 transcript from cases with BCL2 rearrangement.

Of the R-CHOP cohort, 144 cases were evaluated for the presence of t(14;18)(q32;q21) by interphase FISH using dual color “break-apart” probes (Abbott Molecular) as previously described (10). The break-apart signals in more than 5% cells were considered “positive” for translocation. The presence of 3 or 4 signals or more of 18q21 along with 2 signals for centromere-18 was considered as a gain or amplification, respectively.

The χ² test, Spearman correlation coefficient, and Wilcoxon rank-sum test were used to determine the association between categorical variables, BCL2 mRNA/protein, and between BCL2 mRNA/protein and BCL2 rearrangement data, respectively. The Kaplan–Meier method was used for OS and event-free survival (EFS) analysis as done previously (12). Cox regression analysis was used in multivariate modeling of OS and EFS data for both the BCL2 protein and mRNA data, after adjusting for IPI and GEP classification. SAS software V 9.2 (SAS Institute Inc.) was used for all analyses other than creation of the survival trees.

BRB-ArrayTools (version-3.7.0; ref. 31) and gene set enrichment analysis (GSEA; ref. 32) computational programs were used to identify differentially expressed genes and pathways/signatures between BCL2-positive and -negative group in GCB-DLBCL.

We examined BCL2 mRNA expression levels in 221 DLBCL cases in the R-CHOP cohort including GCB-DLBCL (n = 102, 46%), ABC-DLBCL (n = 88, 40%), and unclassifiable DLBCL (n = 31, 14%), with protein expression and BCL2 translocation data available in 169 and 144 cases, respectively. A flowchart outlining the number of patients in each DLBCL subgroups for mRNA, protein, t(14;18) analysis is shown in Fig. 1. The clinical features at the time of presentation for the R-CHOP cohort (n = 221) were not significantly different when compared with the CHOP cohort (n = 180), we had studied previously (12; Supplementary Table S1), except for better OS (P < 0.001) in the R-CHOP cohort. As expected, the cell of origin (P < 0.005) and the IPI (P < 0.0001) were both independent predictors of OS in each cohort (Supplementary Fig. S1).

There were significant differences in some of the clinical features associated with BCL2 protein expression, in DLBCL as a single entity with a higher median age and stage seen in the BCL2-positive group (Table 1), but no differences were observed with regard to mRNA expression (Supplementary Table S2). Among the cell-of-origin subgroups, there were no differences in clinical features in the ABC-DLBCL cases with regard to BCL2 protein expression. However, in the GCB-DLBCL cases, a higher median age, stage, lactate dehydrogenase (LDH) levels, and IPI scores were in the BCL2 protein–positive group (Table 1), whereas no differences were seen with regard to mRNA expression (Supplementary Table S2).

The t(14;18) was observed in 19% (27 of 144) of the DLBCL cases and 34% (22 of 64) of GCB-DLBCL cases, consistent with our previous finding in the CHOP cohort (10). However, 4 cases with the t(14;18) occurred in the ABC-DLBCL. Interestingly, further analysis of these 4 cases showed c-MYC rearrangement (double hit) in 2 and BCL2 amplification (>4 copies) in the other 2 cases.

BCL2 protein expression was observed in 44% (75 of 169) of the DLBCL and was more frequent in ABC-DLBCL (62%; 45 of 73) than in GCB-DLBCL (30%; 22 of 73; P = 0.0002). A similar trend was observed at the mRNA level as well, with significantly higher expression observed in ABC-DLBCL (Table 2). We also observed a significant association between BCL2 mRNA and protein expression in DLBCL as a group and in both the ABC and GCB subgroups (the Spearman correlation 0.64, P < 0.0001). However, there were a number of discrepant cases (3 of 42) in lowest mRNA quartile which showed high (≥50%) protein expression, and 8 (of 43) cases in highest mRNA quartile showed low (<50%) protein expression, that may be due to methodologic limitations or biological variance such as differences in posttranscriptional regulation of BCL2 expression or variations due to stromal components expressing BCL2. The t(14;18) was also significantly associated with an increased BCL2 protein expression (P < 0.0001) and mRNA level (P < 0.0001) in GCB-DLBCL (Supplementary Table S3). However 2 GCB-DLBCL cases with the t(14;18) were negative for protein expression and showed very low mRNA expression, whereas 6 cases without the t(14;18) showed high BCL2 mRNA and protein expression.

The evaluation of BCL2 copy number changes revealed a significant association of 18q21 amplification (>3 copies, P = 0.01) with ABC-DLBCL (12 amplified and 8 gain; of 31 cases) compared with GCB-DLBCL (3 amplified and 4 gain; of 29 cases). Interestingly, all 3 cases of GCB-DLBCL with an amplified 18q21 locus were negative for BCL2 protein expression and had low mRNA expression whereas, in ABC-DLBCL, the majority of cases (10 of 12) with an amplified 18q21 locus were positive for protein expression and showed high mRNA expression (Supplementary Fig. S2).

BCL2 protein expression by immunostaining with 50% cutoff point for positive cases was significantly associated with OS (P = 0.009) or EFS (P = 0.001) in the entire cohort of patients with DLBCL (Fig. 2). To further substantiate this finding, we added 62 DLBCL cases that were entered into the study but did not have acceptable GEP data to the 169 cases with GEP data for a total of 231 cases. Similar results were obtained with for both OS (P = 0.005) and EFS (P < 0.001), thus providing confirmation that BCL2 is a prognostic marker in R-CHOP–treated patients (Supplementary Fig. S2).

Because mRNA expression is a continuous variable, we divided the patients arbitrarily into 2 or 4 groups according to the BCL2 transcript levels. When DLBCL was analyzed as a single entity, patients with high BCL2 mRNA levels had a significantly worse EFS regardless of whether cases were divided into 2 groups (P = 0.006), or quartiles (P = 0.036), but no significant difference was observed in OS (Fig. 3).

When the prognostic significance of BCL2 expression was analyzed according to the DLBCL cell of origin, we observed strikingly different results compared with our previously studied CHOP cohort (12). There is now no significant association of either BCL2 protein or mRNA expression and survival in ABC-DLBCL (Fig. 4), whereas a significant association of OS (P = 0.03) and EFS (P = 0.002) with BCL2 protein expression is now observed in GCB-DLBCL (Fig. 5). At the mRNA level, no significant association was observed with OS or EFS in GCB-DLBCL, but a trend was observed in EFS. This trend was more prominent when GCB-DLBCL cases were divided into quartiles (Supplementary Fig. S3).

The multivariate analysis of OS and EFS in DLBCL as a single entity showed that BCL2 protein was a marginally significant predictor of OS (HR: 2.0, 95% confidence interval (CI), 1.0–4.0; P = 0.06) and a significant predictor of EFS (HR: 2.0, 95% CI, 1.1–3.6; P = 0.02) independent of IP1. However BCL2 mRNA expression was a marginally significant predictor of EFS (P = 0.06) but not OS (P = 0.15). Similar analysis in GCB-DLBCL subgroup showed that protein expression was predictive of EFS (HR: 4.5, 95% CI, 1.2–16.5; P = 0.02) but not OS (HR: 3.0, 95% CI, 0.6–15.8) after adjusting for IPI. However, effective sample size for this analysis was small (n = 59).

More than 500 transcripts were differentially expressed (>1.5-fold and P < 0.005) between the BCL2-positive and -negative cases in the GCB subgroup (Fig. 6). As expected BCL2 transcripts were highly expressed in the BCL2-positive cases, but some proapoptotic genes, for example, BCL2L1(BIM) were also highly expressed. Approximately half of the transcripts upregulated in the BCL2-positive cases were uncharacterized. However, the informative ones included a heterogeneous group that has been recently associated with B-cell neoplasms including FOXP2 (33), CEACAM1 (34), CLLU1 (35), and AKT2 (36). Many of these genes either promote survival or regulate B-cell signaling. Interestingly, a large number of genes overexpressed in the BCL2-negative group were involved in cell adhesion or regulating the extracellular matrix (DSG2, GJB2, CLDN1, NRXN3, PARD3, and CADM1). When the significance level of the t test was lowered to P = 0.01 for differential expression, the majority of genes involved in cell-cycle progression and regulation were upregulated in the BCL2-negative groups, similar to our previous findings (10). In contrast, genes involved in apoptosis (BAD, BAK1, BTG1, and BNIP3L) and BCR signaling (BCAP29, BLK, and LYN) or mainly B-cell related (FCAR, FCRL1, FCRL2) were more prominent in the BCL2-positive group. GSEA identified enrichment of the proliferation signature, dendritic cell (resting signature), DLBCL stromal-1 signature, HIF1-regulated gene signature, and normal mesenchymal signature in the BCL2-negative group (Fig. 6). These observations are consistent with previous findings that the stromal-1 signature is significantly associated with HIF1-α (37), normal mesenchymal signature, and histiocytes and can predict better OS (3). The BCL2-positive group did not show any significantly enriched pathways with the exception of the T_FH signature, suggesting increased infiltration of T_FH cells in these cases.

BCL2 regulates programmed cell death and plays an important role in the response of malignant cells to a variety of stresses that lead to apoptosis, including chemotherapy (38). The association of BCL2 expression with survival in patients with DLBCL treated with CHOP or CHOP-like regimens had conflicting results, with studies showing either significant or no significant association with OS (13–19). We have previously shown that the prognostic significance of BCL2 expression in the context of DLBCL cell of origin, which may explain many of these conflicting findings (12). We and others have shown that patients with ABC-DLBCL treated with CHOP-like therapies have a shorter OS if BCL2 is overexpressed (12, 20), which may be due to the ability of these tumor cells to resist apoptosis induced by chemotherapy. We have also shown mechanistic differences in BCL2 upregulation in the GCB and ABC subgroups (10).

The addition of rituximab to CHOP-like protocols for DLBCL has led to a significant (P < 0.001) increase in patient survival, and such regimens are now considered the standard of care for DLBCL (39, 40). Several recent studies of R-CHOP cohorts have failed to show a BCL2 prognostic effect due to a disproportionate benefit from rituximab of BCL2-positive cases and with no association of BCL2 expression with OS in DLBCL (23–25). Similarly, a patient cohort treated with rituximab and EPOCH also showed no association of BCL2 with OS or EFS (41). However, other studies have shown a prognostic influence of BCL2 mRNA (42) or protein expression in R-CHOP–treated cohorts (43–45). Song and colleagues (22) reported significant correlation of BCL2 protein expression with OS in GCB-like DLBCL, whereas Nyman and colleagues showed significance in non–GCB-DLBCL, only marginal (P = 0.07) in GCB-like DLBCL (43). Another report of R-CHOP cohort has indicated that BCL2 protein expression influences relative risk (RR) in OS (RR = 2.3, P = 0.06) and EFS (RR = 2.2, P = 0.03) in BCL6-positive DLBCL but not in BCL6-negative cases (46), which is significantly associated with non–GCB-DLBCL. The conflicting reports about the prognostic significance of BCL2 expression in the literature can partly be attributed to the following: (i) heterogeneity of the DLBCL cases studied with different proportion of GCB and ABC-DLBCL cases, (ii) patient population with different risk factors other than DLBCL subtype distinction, (iii) variables in management, (iv) technical factors affecting immunostaining, and (v) experience and subjectivity of the pathologist scoring the cases. We had discussed some of these issues in our previous study (12) and suggested that biomarkers should be evaluated in context of molecular subgroups. The heterogeneity within DLBCL has usually been addressed by immunohistochemically defined subgroups, but this approach had not been very consistent in defining prognostic groups among various laboratories due mainly to (iv) and (v) discussed earlier. Different algorithms have also been created by evaluating the expression of several antigens and used as surrogate for GEP-based classification. Interestingly, when we evaluated these algorithms against GEP-defined subtypes, most of these correlated very well with all the immunostains carried out in one laboratory (47), unlike a recent study, where none of the algorithms showed any prognostic significance (48). Because GEP-defined molecular subgroups remain the gold standard for DLBCL classification, we have used GEP-defined subgroups in our study to avoid all the variables from immunohistochemical classification from influencing our results.

In this study, tumors were considered positive, if at least 50% of neoplastic cells show BCL2 expression. The 50% cutoff value has been frequently but not consistently used in assessing BCL2 positivity. The optimal cutoff point was, therefore, chosen by survival tree method (28) as an unbiased approach using P values from adjusted log-rank statistics. A similar approach has been used recently in large DLBCL series for different biomarkers cutoff points (49). Our study shows that when patients with DLBCL are treated with rituximab and CHOP-like therapies, cases who overexpress BCL2 protein have significantly poorer OS (P = 0.009) and EFS (P = 0.001). This association was observed at the mRNA level for EFS as well. Similar to our previous findings in CHOP cohort (12), there were also significant differences in other basic clinical features observed in the BCL2 protein–positive and -negative groups in this R-CHOP series as well, suggesting that BCL2 expression may be associated with age and stage, but these differences were not observed when cases were segregated by mRNA level. Likewise IPI score, Karnofsky score and stage showed significant differences between BCL2 protein–positive and -negative group in GCB-DLBCL but not at mRNA level. Consistent with our previous observations and other reports, BCL2-positive cases are observed more frequent in ABC-DLBCL than in GCB-DLBCL (10, 22, 43). However, the expression of BCL2 mRNA or protein expression was not predictive of either OS or EFS in ABC-DLBCL in our R-CHOP cohort, unlike our CHOP cohort reported previously (12). In contrast, we have now observed a significant association of BCL2 protein expression with poor OS and EFS in GCB-DLBCL, and this was observed at the mRNA level with marginal significance in EFS only (Supplementary Fig. S3). The association of BCL2 with EFS was independent of other clinical features as shown in multivariate analysis. Therefore, we conclude that BCL2-positive ABC-DLBCL has benefited proportionally more from R-CHOP than BCL2-negative tumors, narrowing the differences in survival in this group. However, this is not true for GCB-DLBCL, with BCL2-positive tumors benefiting less from this treatment compared with BCL2-negative tumors, thus resulting in a significant difference in patient survival. Although rituximab was reported to significantly improve the outcome of patients with BCL2-positive DLBCL, a distinction was not made between the subtypes (50).

Our findings raise an important consideration about the interaction of rituximab with the 2 DLBCL cell-of-origin subgroups. In ABC-DLBCL, overexpression of BCL2 protein is associated with constitutive activated NF-κB pathway (11) and 18q21 amplification (12). In vitro experiments have shown that rituximab downregulates NF-κB and its target BCL2 (51), and it may be similarly effective in reducing the expression of BCL2 in vivo, resulting in increased susceptibility to chemotherapy. On the other hand, rituximab may fail to downmodulate BCL2 in GCB-DLBCL, which contributes to drug resistance in this subset of tumors. If this is indeed correct, then additional measures that reduce BCL2 levels or function may be effective in improving survival in this group of patients with GCB-DLBCL. BCL2 may also interact with other oncogenic pathways as shown by the consistently poor prognosis in patients with high expression of both BCL2 and c-MYC genes (52). Because BCL6 is almost universally expressed by GCB-DLBCL and can repress genes that control DNA damage response (53) and checkpoint p53, ATR (54, 55), it may synergize with high BCL2 expression in resisting drug-induced apoptosis in GCB-DLBCL.

One unexpected observation in our series was that the t(14;18) was not predictive of OS or EFS in GCB-DLBCL, even though it was highly associated with BCL2 expression. This may be due to subthreshold (<50%) protein expression in 36% of the t(14;18)-positive cases and the expression of BCL2 protein (>50%) in 8% of the GCB-DLBCL cases lacking this translocation, thus reducing the correlation between BCL2 translocation and survival.

To investigate further the mechanisms that may play a role in the difference in survival between the BCL2-positive and -negative GCB-DLBCL groups, we analyzed the differences in GEP between these 2 groups. We found that the stromal-1 signature was associated with the BCL2-negative group (3). The “stromal-1” signature is related to extracellular matrix deposition and mesenchymal and histiocytic cell infiltration and is associated with favorable outcome (3). The BCL2-negative group was also associated with higher proliferation, confirming what we observed previously (10). The BCL2-positive group did not show any significantly enriched gene expression signatures, except T_FH cell–related signatures, which suggest the presence of increased infiltrating of T_FH cells. A recent observation showed that adhesion of lymphoma cells to follicular dendritic cells can protect the neoplastic cells against apoptosis or promote drug resistance by inducing miR-181a expression resulting in downregulating BIM protein level (56). Similar protective signal may be present, as T_FH cells are often associated with follicular dendritic cells. BIM downregulation and BCL2 expression may provide synergistic resistance to therapy.

In summary, our study have shown the importance of reevaluating prognostic factors in the setting of new therapeutic regimens and provided a new perspective in BCL2 expression and prognosis with respect to the cell-of-origin classification of DLBCL. Our findings suggest improvement in outcome on the use of rituximab with CHOP in ABC-DLBCL and the BCL2 protein–negative subset of GCB-DLBCL. However, the BCL2-positive subset of GCB-DLBCL has shown less improvement, and these cases may benefit from novel agents such as inhibitors of BCL2 function.

No potential conflicts of interest were disclosed.

The authors thank Martin Bast for clinical data collection and Cynthia M. Lachel and Greg Cochran for technical assistance.

J. Iqbal and W.C. Chan designed, conducted study, supervised all aspects of the research and analysis, and wrote and finalized the manuscript. P.N. Meyer, D.D. Weisenburger, and T.C. Greiner were responsible for design of the study, pathology review, and scoring immunohistochemical stains and for the final approval of the manuscript. L.M. Smith was responsible for the statistical analysis section in the study. N.A. Johnson conducted FISH and assisted in manuscript writing. J.M. Connors, L.M. Staudt, L. Rimsza, E. Jaffe, A. Rosenwald, G. Ott, E. Campo, R.M. Braziel, J.R. Cook, R.R. Tubbs, and R.D. Gascoyne were involved in the design of the study, pathology review, and provided the DLBCL cases for the study. J.M. Vose, J.M. Connors, and J.O. Armitage reviewed the clinical aspects and writing of manuscript.

This work was supported in part by NCI grant 5U01/CA114778. This study is part of Lymphoma/Leukemia Molecular profiling project (LLMPP).

The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.