Prognostic impact of B-cell lymphoma 6 in primary CNS lymphoma

Abstract

Background

We investigated the prognostic significance of B-cell differentiation status and common B-cell differentiation markers in a post hoc analysis of 119 patients with primary CNS lymphoma (PCNSL) homogeneously receiving high-dose methotrexate (HDMTX)–based chemotherapy within the prospective G-PCNSL-SG1 trial.

Methods

We evaluated protein expression of B-cell lymphoma 2 (BCL2), BCL6, CD10, and multiple myeloma oncogene 1/interferon regulatory factor 4 (MUM1/IRF4) by immunohistochemistry and analyzed the association with survival.

Results

The median follow-up of all patients was 67.5 months. Median progression-free survival (PFS) was 10.61 months (95% CI: 4.23–17.00). Median overall survival (OS) was 28.85 months (95% CI: 17.96–39.73). Eighty-nine tumors expressed BCL2 (92.7%), 24 (20.5%) expressed CD10, 60 (54.1%) expressed BCL6, and 87 (79.0%) expressed MUM1/IRF4. On the basis of the Hans algorithm, 80 tumors (73.4%) were classified to the non–germinal center B group, suggesting a post–germinal center origin of PCNSL. Expression of BCL6 (cutoff point 30%), but none of the other markers, was associated with shorter PFS (P = .047) and OS (P = .035). On multivariate analysis, BCL6 expression was associated with shorter PFS (hazard ratio: 1.95, 95% CI: 1.22–3.12, P = .005) but not OS (hazard ratio: 1.85, 95% CI: 0.71–4.80, P = .21). Classification according to Hans algorithm and expression status of the single B-cell markers BCL2, CD10, and MUM1/IRF4 did not correlate with prognosis.

Conclusion

The findings are limited by the fact that only 23% of all G-PCNSL-SG1 patients could be included in the analysis. If validated in an independent cohort, BCL6 may assume clinical relevance as an unfavorable prognostic biomarker in PCNSL.

Primary CNS lymphoma (PCNSL) is a rare aggressive B-cell neoplasia most frequently of diffuse large B-cell lymphoma (DLBCL) type that is confined to the CNS at time of diagnosis.¹ The formerly poor prognosis of PCNSL has been greatly improved by systemic high-dose methotrexate (HDMTX)–based polychemotherapy, which provides the current standard treatment for all patients with PCNSL. Despite substantial improvements in the treatment of PCNSL, the response to HDMTX-based chemotherapy is quite heterogeneous and overall prognosis remains poor, with a median progression-free survival (PFS) of ∼12 months and a median overall survival (OS) of about 3 years in most studies. This necessitates the identification and evaluation of reliable predictive and prognostic biomarkers for risk-stratified treatment decisions.²

In systemic DLBCL, the prognostic value of different biologic markers has been widely studied. Moreover, at least 2 prognostically important subgroups (germinal center B-cell-like and activated B-cell-like DLBCL) were characterized by gene expression profiling using a cDNA microarray.^3,4 Due to the impracticability to perform microarray analysis on every patient, various immunohistochemical (IHC) algorithms have been developed to capture the molecular diversity and stratify patients according to survival.^5–8 The most widely used method is the Hans algorithm separating DLBCL into germinal center B cell (GCB) and non–germinal center B cell (non-GCB) groups by using antibodies against cluster of differentiation (CD)10, B-cell lymphoma 6 (BCL6), and multiple myeloma oncogene 1/interferon regulatory factor 4 (MUM1/IRF4).⁹ In PCNSL, a few small and mostly retrospective studies either failed to observe a prognostic impact of common B-cell differentiation markers or revealed contradictory results.^10–14 Here we present the analysis of a set of B-cell differentiation markers and their prognostic impact in a cohort from the G-PCNSL-SG1 trial, the largest trial ever conducted in PCNSL.

Materials and Methods

Patients and Treatment

Included in our study were immunocompetent adult patients with newly diagnosed PCNSL in the G-PCNSL-SG1 trial with tumor samples available at one of the 2 reference pathology centers and tumor amount sufficient for IHC analysis. The inclusion and exclusion criteria were reported previously.¹⁵ Patients were randomly allocated to receive first-line HDMTX-based chemotherapy with or without subsequent whole-brain radiotherapy (WBRT). Between May 2000 and August 2006, study therapy consisted of HDMTX (4 g/m² as a 4-h i.v. infusion with dose reduction according to creatinine clearence) on day 1 of six 14-day cycles; thereafter, patients were to receive HDMTX plus ifosfamide (1.5 g/m²) on days 3–5 of six 14-day cycles. In those assigned to receive first-line chemotherapy followed by radiotherapy, WBRT was to be given at a total dose of 45 Gy in 1.5-Gy fractions. Patients allocated to first-line chemotherapy without WBRT who had not achieved complete response to HDMTX-based chemotherapy were given high-dose cytarabine (2 × 3 g/m²) on days 1–2 of 22-day cycles.

The study protocol was approved by local institutional review boards or ethics committees. All participants gave written informed consent.

Immunohistochemical Analysis

Immunostaining on paraffin sections was performed centrally at 2 study reference pathology centers for CD10 (clone 56C6, Leica Biosystems), BCL6 (clone LN22, Leica Biosystems; clone PG-B6p, Dako), BCL2 (clone100/D5, Leica Biosystems; clone 124, Dako), and MUM1/IRF4 (clone MUM1p, Dako) using an automated immunostainer (BondMax, Leica Biosystems; Benchmark XT, Roche-Ventana). Antigen retrieval was performed using appropriate conditions, bound antibodies were made visible employing a Bond Polymer Refine DAB or ultraView Universal DAB detection kit (Roche-Ventana), developed and counterstained using the manufacturer′s protocols and reagents.

Immunohistological labeling of BCL2, CD10, BCL6, and MUM1/IRF4 was visually analyzed and interpreted according to the methods and cutoff points previously published by Horn and colleagues.¹⁶ For BCL6 staining, tumors with more than 30% positively labeled cells were considered positive. Tumors were further classified into GCB and non-GCB subgroups according to their expression of CD10, BCL6, and MUM1/IRF4 using the methods and cutoff points (>30%) published by Hans et al.⁹ Tumors with CD10 expression or BCL6 expression without expression of MUM1/IRF4 were defined as GCB. The remaining tumors were classified as non-GCB.

Statistics

Progression-free survival was defined as the time from study entry to first progression or death from any cause. Overall survival was defined as the time from study entry to death. PFS and OS were estimated by the Kaplan–Meier method. Group comparisons were carried out using the log-rank test. Additionally, simple and multiple Cox proportional hazards regression models were calculated, and hazard ratios (HRs) including 2-sided 95% CIs were calculated. The variables included in the multiple Cox models were determined by forward and backward variable selection leading to identical models. Nonsignificant variables were not included in the final models. Distribution of patients' characteristics to different groups was analyzed by the chi-square test. Mean values of independent groups were compared with Student's t-test. The level of significance was .05 (two-sided). Commercially available software was used (SPSS for Windows, release 21.0).

Results

Patient Characteristics and Clinical Outcome

As central collection of tumor tissue was not intended at the time of study enrollment, all available tumor material from participating centers was collected. Of 526 eligible patients treated in the G-PCNSL-SG1 trial, sufficient tumor material of 119 patients was available for the present IHC analysis. The main patient characteristics, treatment, and outcome are summarized in Table 1. With the exception of multifocal brain involvement and type of initial surgery for PCNSL, patient characteristics were comparable to those of study patients not included in this analysis (Supplementary Table S1). However, there was a nonsignificant trend to better PFS (P = .057) and OS (P = .056) in patients included in our IHC analyses.

Table 1.

Patient characteristics, treatment, and treatment response

Characteristics	All Patients, N = 119; n (%)
Median age, y (range)	63 (26–83)
Males	62 (52.1)
KPS, median	70
MSKCC score
1	13 (10.9)
2	59 (49.6)
3	31 (26.1)
No specification	16 (13.4)
Serum LDH elevated	23 (19.3)
No specification	59 (49.6)
Meningeal involvement	6 (5.0)
No specification	30 (25.2)
Number of cerebral lesions
0–1	82 (68.9)
≥2	30 (25.2)
No specification	7 (5.9)
HDMTX vs HDMTX/IFO	89/30
WBRT	75 (63.0)
Response to HDMTX-based chemotherapy
Complete response	53 (44.5)
Partial response	20 (16.8)
Stable disease	6 (5.0)
Progressive disease	27 (22.7)
No specification	13 (10.9)
Median PFS, mo (95% CI)	10.61 (4.23–17.00)
Median OS, mo (95% CI)	28.85 (17.96–39.73)

Abbreviations: LDH, lactate dehydrogenase; IFO, ifosfamide.

The median follow-up of all patients studied here was 67.5 months. The median PFS was 10.6 months (95% CI: 4.2–17.00); the median OS was 28.9 months (95% CI: 18–39.7).

Immunohistochemical Profiles

Eighty-nine of 96 tumors (92.7%) expressed BCL2 and 24 of 117 tumors (20.5%) expressed CD10. BCL6 and MUM1/IRF4 were expressed in 60 of 111 (54.1%) and 87 of 110 (79.0%) tumors, respectively. Fifty-three tumors of the BCL6-positive tumors coexpressed MUM1/IRF4 (88.3%). Twenty-nine tumors (26.6%) were classified as GCB and 80 (73.4%) as non-GCB.

Prognostic Analysis

Among the biologic markers, only BCL6 expression correlated with shorter PFS and OS (P = .047 and P = .035, respectively; Fig. 1). On univariate analysis, the association of BCL6 was significant for both PFS (HR: 1.53, 95% CI: 1.01–2.34, P = .047) and OS (HR: 1.66, 95% CI: 1.04–2.65, P = .035). On multivariate analysis, however, a significant association with PFS only was found (HR: 1.95, 95% CI: 1.22–3.12, P = .005) (Table 2). No other biologic markers tested correlated with outcome. Particularly, classification according to the Hans algorithm revealed no significant difference between GCB and non-GCB subgroups with respect to survival outcome.

Fig. 1.

Comparison of PFS and OS according to BCL6 expression.

Table 2.

Univariate and multivariate analyses for PFS and OS

	Univariate Analysis			Multivariate Analysis
	HR	P	95% CI	HR	P	95% CI
PFS
Agea	1.01	.32	0.99–1.03	n.d.	–	–
KPSb	0.99	.23	0.98–1.01	n.d.	–	–
Genderc	0.81	.29	0.54–1.20	0.69	.108	0.44–1.08
Multifocal brain involvementd	1.79	.011	1.14–2.80	1.20	.49	0.71–2.03
MSKCC scoree	1.40	.14	0.89–2.20	1.87	.011	1.15–3.04
LDHf	1.27	.43	0.70–2.28	1.41	.32	0.72–2.76
Surgery (biopsy vs resection)	1.28	.24	0.85–1.91	0.94	.80	0.57–1.54
HDMTX vs HDMTX/IFO	0.99	.97	0.62–1.58	1.00	1.00	0.60–1.67
BCL2g	1.28	.60	0.52–3.17	1.09	.86	0.43–2.77
BCL6h	1.53	.047	1.01–2.34	1.95	.005	1.22–3.12
CD10i	0.81	.42	0.49–1.34	0.82	.46	0.47–1.40
MUM1/IRF4j	0.96	.89	0.58–1.60	0.77	.37	0.43–1.37
GCB/non-GCBk	1.12	.65	0.70–1.80	1.20	.49	0.72–2.00
OS
Agea	1.027	.26	1.003–1.05	n.d.	–	–
KPSb	0.99	.44	0.97–1.00	n.d.	–	–
Genderc	0.75	.20	0.49–1.16	0.48	.091	0.20–1.13
Multifocal brain involvementd	1.47	.11	0.91–2.39	2.72	.019	1.18–6.28
MSKCC scoree	1.66	.041	1.02–2.70	2.95	.016	1.22–7.13
LDHf	1.10	.77	0.59–2.05	1.89	.38	0.46–7.80
Surgery (biopsy vs resection)	1.17	.50	0.75–1.82	0.82	.67	0.33–2.02
HDMTX versus HDMTX/IFO	0.92	.76	0.53–1.60	0.37	.19	0.084–1.63
BCL2g	0.94	.89	0.38–2.33	0.68	.62	0.16–3.01
BCL6h	1.66	.035	1.04–2.65	1.85	.21	0.71–4.80
CD10i	0.61	.09	0.35–1.07	1.00	1.00	0.39–2.57
MUM1/IRF4j	1.00	.99	0.57–1.77	2.13	.32	0.48–9.41
GCB/non-GCBk	1.46	.17	0.86–2.48	1.14	.79	0.44–2.98

Abbreviations: n.d., not determined; LDH, lactate dehydrogenase; IFO, ifosfamide.

^aAge ≥60 y vs <60 y.

^b10% increase in KPS. In multivariate analyses, age and KPS were not analyzed separately but combined as the MSKCC score.

^cFemale vs male sex.

^d≥2 lesions vs 0–1 cerebral lesions.

^eMSKCC score 3 vs 1/2.

^fElevated vs normal LDH.

^gBCL2-positive vs negative expression.

^hBCL6-positive vs negative expression.

ⁱCD10-positive vs negative expression.

^jMUM1/IRF4-positive vs negative expression.

^kGCB vs non-GCB according to Hans algorithm.

Patient characteristics and response to treatment did not differ between BCL6-positive and -negative patients except for number of brain lesions (Supplementary Table S2), with BCL6 expression being significantly associated with multifocal brain involvement (BCL6 positive: 33.3% vs BCL6 negative: 15.7%).

When the impact of commonly accepted clinical risk factors in PCNSL was tested, a significant association of age (HR: 1.027, 95% CI: 1.003–1.05, P = .026) and Karnofsky performance score (KPS) (HR: 0.99, 95% CI 0.97–1.00, P = .044) with OS was found on univariate analysis. The Memorial Sloan Kettering Cancer Center (MSKCC) score, which is deduced from age and KPS, showed in multivariate analyses a prognostic association with both PFS (HR: 1.87, 95% CI: 1.15–3.04, P = 0.011) and OS (HR: 2.95, 95% CI: 1.22–7.13, P = .016). Multifocal brain involvement was significantly associated with shorter PFS (HR: 1.79, 95% CI: 1.14–2.80, P = .011) on univariate and shorter OS (HR: 2.72, 95% CI: 1.18–6.28, P = .019) on multivariate analysis. To exclude that the effect of multifocal disease was due to confounding with the type of surgery, we performed adjusted and stratified analysis for these 2 factors for OS and PFS. In both adjusted analyses, multifocal disease remained significant after inclusion of type of surgery: PFS odds ratio = 1.78, P = .012; OS odds ratio = 2.77, P = .012. In cases with resection, the odds ratio for PFS was 2.06 (P = .043) and for OS 2.13 (P = .23); in cases with biopsy, the odds ratio for PFS was 1.58 (P = .131) and for OS 2.97 (P = .044) (the lack of significance in some subgroups was most probably due to their small size).

Discussion

Our data confirm previous studies indicating an activated B-cell-like immunophenotype and post–germinal center origin of most PCNSL^12,17–19 with frequent expression of the activation marker MUM1/IRF4, infrequent CD10 expression, and expression of BCL6, a marker of germinal center, in about half of the cases. In normal B cells, BCL6 and MUM1/IRF4 are exclusively expressed,²⁰ and a high percentage of BCL6/MUM1/IRF4 coexpression has been shown to be a characteristic feature of PCNSL compared with systemic DLBCL.^9,12,21–23 Confirming previous analyses, the vast majority of our BCL6-positive tumors coexpressed MUM1/IRF4, which indicates the activated immunophenotype of PCNSL.^12,21 Additionally, most PCNSL of our series belonged to the non-GCB group on the basis of the Hans algorithm, which is in accordance with earlier IHC analyses^12,17–19 and gene expression profiling analysis of 21 PCNSL cases.²⁴

The prognostic utility of B-cell differentiation status and various B-cell differentiation markers to predict outcome in PCNSL patients is currently questionable. The available data concerning this issue are based on mostly retrospective studies with small patient numbers and heterogeneous treatment schedules.^{10–14,18,19,21,25,26} We here investigated expression and prognostic significance of B-cell differentiation markers in the so far largest cohort of PCNSL patients homogeneously receiving HDMTX-based chemotherapy in the prospective G-PCNSL-SG1 trial. BCL2 as an anti-apoptotic protein and normally downregulated in GCB has been shown to be an independent poor prognostic indicator for systemic DLBCL.^27,28 In our series, we did not observe a correlation of BCL2 expression with PFS or OS. Considering the low frequency of BCL2-negative PCNSL, BCL2 expression seems to be more a characteristic than a prognostic feature in PCNSL. The expression status of the single B-cell markers CD10 and MUM1/IRF4 also did not correlate with prognosis, which is in accordance with previous analyses by Braaten et al²⁶ and others.^12,14 However, the percentage of CD10-positive tumors might have been too low, increasing the potential for confounding variables.

Previous analyses of the prognostic significance of BCL6 expression in PCNSL yielded controversial results. Lossos et al¹³ reported an independent positive association with PFS (and nearly significant with OS), and Braaten et al²⁶ found an independent and significant positive association with OS. Both were retrospective studies with heterogeneously treated patients. Moreover, an unusually low positivity threshold of only 10% was used by Braaten et al, resulting in an expression frequency of 79%. In contrast, Momota et al¹⁴ found an independent negative association of BCL6 expression with PFS. In the only prospective multicenter trial addressing this question (CALGB 50202), BCL6 expression significantly correlated with shorter survival.¹¹ Our findings confirm a negative correlation of BCL6 expression with outcome in PCNSL. Reasons for discrepancy with other studies could be the retrospective nature of previous studies, few patient numbers with consequent limited statistical power and possibility of confounding variables, non-uniform treatment, and different immunophenotype and classification methods among different studies. Interestingly, there seems to be a correlation between BCL6 expression and negative clinical characteristics in PCNSL. Camilleri-Broet et al¹² found an association with higher age and worse performance status. In the present analysis, a correlation between BCL6 expression and multifocal brain involvement was found. Multifocal brain involvement and inferior survival outcome were findings in the whole G-PCNSL-SG1 cohort²⁹ and have been reported by others.³⁰

In our series, we did not observe a significant prognostic difference between the GCB and non-GCB subgroups. This is in line with several smaller studies.^12,14,18,25

Our study has several limitations that have to be regarded. Although based on the randomized controlled G-PCNSL-SG1 trial, the presented data are a formerly unplanned post hoc analysis of only 23% of all G-PCNSL-SG1 patients with an—unintended—positive selection bias for analyzed patients. The bias was due to selection of patients on the basis of availability of large tumor amount, which resulted in a higher percentage of patients with <2 tumor lesions and consecutively subtotal or total tumor resection. These factors were associated with better outcome in the total G-PCNSL-SG1 study population²⁹ and might explain the nonsignificant but potentially biologically important trend to a superior prognosis in patients included in our IHC analyses. It remains unclear how this bias might have impacted our findings. Further shortcomings of the study are the number of patients analyzed and the number of events (PFS: 99, OS: 83). Although high compared with previous studies, both might have been too low for some outcome differences to become significant, and the multivariate analysis has to be interpreted with caution. Even though the selection of variables was identical with forward and backward methods, a change of values in a small number of patients might have led to different models. This may also explain the differences between univariate and multivariate analyses for several factors that showed substantially higher odds ratios in multivariate compared with univariate analysis (eg, MSKCC score and multifocal brain involvement in the analysis of OS).

Most of our patients (74.7%) were treated with HDMTX alone, which was the standard first-line treatment regimen at the time of study initiation. The type of chemotherapy did not have a significant prognostic impact in our uni- and multivariate analyses, but the number of patients might have been too small. Thus, we cannot exclude that the survival difference between the BCL6-positive and BCL6-negative groups within our investigation could have been due to the less effective treatment regimen and therefore might not be applicable for patients undergoing an intensified first-line treatment.

In conclusion, our findings confirm an activated B-cell-like immunophenotype and post–germinal center origin of PCNSL and indicate BCL6 expression as a valuable biomarker for inferior prognosis. In view of the fact that several previous studies reported contradicting results, further prospective studies are necessary to validate our results.

Funding

The G-PCNSL-SG1 trial was supported by German Cancer Aid. This research received no grant from any funding agency.

Conflict of interest statement. The authors report no potential conflicts of interest.