Gene Expression Signature–Based Prognostic Risk Score in Patients with Primary Central Nervous System Lymphoma

Published OnlineFirst August 20, 2012; DOI: 10.1158/1078-0432.CCR-12-0596

Clinical Cancer Imaging, Diagnosis, Prognosis Research

Atsushi Kawaguchi1, Yasuo Iwadate2, Yoshihiro Komohara3, Masakazu Sano4, Koji Kajiwara5, Naoki Yajima4, Naoto Tsuchiya4, Jumpei Homma4, Hiroshi Aoki4, Tsutomu Kobayashi4, Yuko Sakai7, Hiroaki Hondoh6, Yukihiko Fujii4, Tatsuyuki Kakuma1, and Ryuya Yamanaka7

Abstract Purpose: Better understanding of the underlying biology of primary central nervous system lymphomas (PCNSL) is critical for the development of early detection strategies, molecular markers, and new therapeutics. This study aimed to define genes associated with survival of patients with PCNSL. Experimental Design: Expression profiling was conducted on 32 PCNSLs. A gene classifier was developed using the random survival forests model. On the basis of this, prognosis prediction score (PPS) using immunohistochemical analysis is also developed and validated in another data set with 43 PCNSLs. Results: We identified 23 genes in which expressions were strongly and consistently related to patient survival. A PPS was developed for overall survival (OS) using a univariate Cox model. Survival analyses using the selected 23-gene classifiers revealed a prognostic value for high-dose methotrexate (HD-MTX) and HD- MTX–containing polychemotherapy regimen–treated patients. Patients predicted to have good outcomes by the PPS showed significantly longer survival than those with poor predicted outcomes (P < 0.0001). PPS using immunohistochemical analysis is also significant in test (P ¼ 0.0004) and validation data set (P ¼ 0.0281). The gene-based predictor was an independent prognostic factor in a multivariate model that included clinical risk stratification (P < 0.0001). Among the genes, BRCA1 protein expressions were most strongly associated with patient survival. Conclusion: We have identified gene expression signatures that can accurately predict survival in patients with PCNSL. These predictive genes should be useful as molecular biomarkers and they could provide novel targets for therapeutic interventions. Clin Cancer Res; 18(20); 5672–81. 2012 AACR.

Introduction tion is often diffuse and multifocal, and most frequently A primary central nervous system lymphoma (PCNSL) is affects the supratentorial brain parenchyma, with periven- an extranodal form of non-Hodgkin lymphoma arising in tricular lesions involving the corpus callosum, basal the craniospinal axis. For many years, PCNSLs were ganglia, or thalamus. The absence of systemic lymphade- reported to represent 3% to 5% of all primary central nopathies and other extracranial localizations of disease nervous system (CNS) tumors (1). However, PCNSL seems should be conﬁrmed. Most PCNSLs belong to the diffuse to be increasing in incidence (2–4). The tumor manifesta- large B-cell lymphomas (DLBCL) but differ from systemic DLBCLs by their less favorable prognosis. The systemic use of high-dose methotrexate (HD-MTX)– 1 Authors' Afﬁliations: Biostatistics Center, Kurume University, Kurume, based chemotherapy with radiotherapy for newly diag- Fukuoka; 2Department of Neurosurgery, Graduate School of Medical Sciences, Chiba University, Chiba; 3Department of Cell Pathology, Grad- nosed PCNSL has improved the median overall survival uate School of Medical Sciences, Kumamoto University, Kumamoto; (OS) from 20 to 36 months (5–8). However, there are still 4 Department of Neurosurgery, Brain Research Institute, Niigata University, many individual variations within the diagnostic and prog- Niigata; 5Department of Neurosurgery, Graduate School of Medical Sciences, Yamaguchi University, Ube, Yamaguchi; 6Department of Neu- nostic categories, resulting in a need for additional biomar- rosurgery, Toyama Prefectural Central Hospital, Toyama; and 7Graduate kers, partly because of the inability to recognize these School for Health Care Science, Kyoto Prefectural University of Medicine, Kyoto, Japan patients prospectively. Although, the clinical scoring model using age, Karnofsky performance status (KPS), and lactate Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/). dehydrogenase (LDH) level has prognostic value for PCNSL (9–11), it has not been used successfully to stratify patients Corresponding Author: Ryuya Yamanaka, Kyoto Prefectural University of Medicine, Graduate School for Health Care Science, 465 Kajii-cho, Kami- for therapeutic trials. Molecular markers could improve the gyoku, Kyoto 602-8566, Japan. Phone: 81-75-212-5429; Fax: 81-75-212- outcome prediction, discover potential targets for therapeu- 5423; E-mail: [email protected] tic intervention, and elucidate mechanisms that result in doi: 10.1158/1078-0432.CCR-12-0596 resistance to chemotherapy. A comprehensive molecular 2012 American Association for Cancer Research. approach to predict the prognosis is awaited. In the present

5672 Clin Cancer Res; 18(20) October 15, 2012

Biomarker Study for PCNSL

These tissues contained more than 95% tumor cells. The Translational Relevance quality of the obtained RNA was veriﬁed using a Bioanalyzer In this study, we report the development and valida- System (Agilent Technologies) and RNA Pico Chips (Agilent tion of a risk-score model based on the expression of 23 Technologies). Subsequently, 1 mg of RNA was processed for genes. This 23-gene risk score is highly associated with hybridization to a GeneChip Human Genome U133 Plus the outcome of patients with newly diagnosed primary 2.0 Expression Array (Affymetrix Inc.), which contained central nervous system lymphoma (PCNSL). These approximately 47,000 genes. After hybridization, the chips results suggest the importance of this multimarker panel were processed using a Fluidics Station 450, High-Resolu- as a stratiﬁcation factor for the design of future compar- tion Microarray Scanner 3000, and GCOS Workstation ative therapeutic trials. Version 1.3 (Affymetrix Inc.).

Validation of differential expression by real-time qPCR Quantitative PCR (qPCR) was conducted using a Ste- study, we carried out an expression profiling analysis in pOne Real-Time PCR System (Applied Biosystems) and patients with PCNSL for the identification of genes that are TaqMan Universal PCR Master Mix (Applied Biosystems) predictive of OS. according to the manufacturer’s protocol. The Assays-on- Demand probe/primer sets (Applied Biosystems) used Materials and Methods were as follows: ATAD1, Hs00907773_g1; BRCA1, Samples and study population Hs01556193_m1; FANCA, Hs01116668_m1; GAPDH, Patients were diagnosed and treated at Niigata University Hs99999905_m1; GGH, Hs00914163_m1; GNASAS, Hospital (Niigata, Japan), Chiba University Hospital (Chiba, Hs00294858_m1; PGAM1, Hs01652468_g1; PPP3R1, Japan), Yamaguchi University Hospital (Ube, Yamaguchi, Hs01547793_m1; RBBP8, Hs00161222_m1; ROCK1, Japan), and Toyama Prefectural Central Hospital (Toyama, Hs01127699_m1; STIL, Hs00161700_m1; TRMT6, Japan) between 2000 and 2010. Clinical data were obtained Hs00210942_m1; and ZNF681, Hs01862022_s1. Total through a registered database and chart review. Inclusion RNA (1 mg) was reverse-transcribed into cDNA using Super- criteria were a histology-proven CNS lymphoma without the Script II (Invitrogen), and 1 mL of the resulting cDNA was evidence of systemic lymphoma, and no evidence of HIV-1 used for qPCR. Validation was conducted on a subset of infection, opportunistic infections, or other immunodefi- tumors that were part of the original tumor data set assessed. ciency. Patients were selected on the basis of the availability Assays were carried out in duplicate. The raw data produced of tumor specimens without regard to the clinical outcome. by qPCR referred to the number of cycles required for All patients underwent brain imaging with either computed reactions to reach the exponential phase. Expression of tomography (CT) or magnetic resonance imaging (MRI). glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was After the diagnostic biopsy, detailed history and physical used for normalization of the qPCR data. The mean expres- examination, complete blood count, screening blood tests of sion fold change differences between tumor groups were DDC hepatic and renal function, serum protein electrophoresis, calculated using the 2 T method (12). and chest radiographs were obtained. The CT or MRI of the thorax, abdomen, and pelvis were conducted for all patients. Immunohistochemistry Ophthalmologic consultations and slit-lamp examinations Three antibodies for immunophenotype determination were used to rule out ocular involvement. Bone marrow and 5 commercially available antibodies for proteins biopsy wasnot routinely conducted unless CNS involvement encoded by genes associated with patient survival were was part of a systemic lymphoma. Lumbar puncture for selected for immunohistochemistry (IHC). Sections (5 mm) cerebrospinal fluid (CSF) evaluation was routinely con- of the formalin-fixed, paraffin-embedded tissue specimens ducted. Tissues were snap-frozen in liquid nitrogen within were evaluated. The primary antibodies recognized BCL6 5 minutes of harvesting, and stored at 80 C thereafter. All (DAKO; 1:200 dilution), BRCA1 (Abcam; 1:200 dilution), specimens were centrally reviewed by a board-certified CD10 (Nichirei; 1:1 dilution), CD79a antibody (DAKO; pathologist by observation of sections of paraffin-embedded 1:50 dilution), FANCA (Abcam; 1:3,000 dilution), MUM1 tissues that were adjacent or in close proximity to the frozen (DAKO; 1:50 dilution), PPP3R1 (Abcam; 1:100 dilution), sample from which the RNA was subsequently extracted. The ROCK1 (Sigma–Aldrich; 1:125 dilution), and RBBP8 cut-offs for normal CSF protein and serum LDH levels were (Abnova; 1:200 dilution). Anti-goat or anti-mouse second- 45 mg/dL and 216 IU/L, respectively. Informed consent was ary antibodies (Nichirei) were also applied. The staining obtained from all patients for the use of their samples, in intensity was classified as none or weakly positive (0 points), accordance with the guidelines of the respective Ethical moderately positive (1 point), or strongly positive (2 Committees on Human Research. points). The cases in which positive cells were more than 20% of lymphoma cells were determined to be positive. The RNA extraction and array hybridization averages of 3 independent measurements were calculated Approximately, 100 mg of tissue from each tumor was to the first decimal place. The observers were not aware of subjected to total RNA extraction using Isogen (Nippon the case numbers. For double-immunostaining, after Gene) in accordance with the manufacturer’s instructions. immunostaining of BRCA1 was conducted, sections were

www.aacrjournals.org Clin Cancer Res; 18(20) October 15, 2012 5673

Kawaguchi et al.

washed with glycine buffer (pH 2.2). Then sections were Table 1. Patient characteristics rereacted with anti-CD79a antibody and visualized with HistoGreen (LinarisBiologische). Training set Validation set (n ¼ 32) (n ¼ 43) Bioinformatics analysis The primary outcome was OS defined as the time from Characteristic N (%) N (%) P first diagnosis to death or last follow-up. All statistical Age, y 0.949 analyses were conducted using R software (13) and Bio- Average 64.18 64.34 conductor (14). The Affymetrix GeneChip probe-level data Range 44–76 17–84 were preprocessed using MAS 5.0 (Affymetrix Inc.) for Gender 0.141 background adjustment and log-transformation (base 2). Male 17 (53) 30 (70) Each array was normalized by applying a quantile normal- Female 15 (47) 13 (30) ization to impose the same empirical distribution of inten- KPS at diagnosis 0.392 sities to each array. Genes that passed the filter criteria below Median 70 70 were considered for further analysis. To select predictors 70 19 (59) 24 (56) (genes) for OS, we first set filtered gene expressions and 70> 13 (41) 19 (44) applied the random survival forests-variable hunting (RSF- No. of lesions 0.647 VH) algorithm (15). Among the parameters in the algo- Single 21 (66) 26 (60) n rithm, the number of Monte Carlo iterations ( rep) and Multiple 11 (34) 17 (40) value controlling the step size used in the forward process Deep lesions <0.05 n n ¼ n ¼ ( step) were set as rep 100 and step 5 following Yes 13 (41) 30 (70) Ishwaran and colleagues (15). For other parameters, such No 19 (59) 13 (30) as the number of trees and number of variables selected Histology randomly at each node, we used the default settings in the DLBCL 32 (100) 43 (100) varSelfunction within the RandomSurvivalForest package Other 0 (0) 0 (0) before the selection. We classified the samples into 2 sur- Chemotherapy 0.14 vival groups by a Ward minimum variance cluster analysis, HD-MTX 16 (50) 17 (40) with inputs of ensemble cumulative hazard functions for Polychemo 16 (50) 23 (53) each individual for all unique death time points estimated Other 0 (0) 3 (7) from the fitted random survival forests model to selected genes. NOTE: Polychemo, HD-MTX–containing polychemotherapy. The 2 classified survival groups were used to compute the prognosis prediction score (PPS) from a simple form (linear combination of gene expressions). To do this, we used the years). Seventeen patients (53%) were males and 15 (47%) principal component analysis and receiver operating char- were females. The median preoperative KPS was 70. All the acteristic analysis. Briefly, we computed the first principal patients had histology-proven DLBCL. Twenty-one patients component of the gene expressions selected by the RSF-VH (66%) had a single lesion and 11 (34%) had multiple algorithm as a risk score, and then searched for the optimal lesions. Deep structures of the brain, that is, the periven- cut-off point to predict survival groups with maximum tricular lesion, basal ganglia, corpus callosum, brain stem, accuracy by the Yoden index (16). The validation for this and/or cerebellum, were involved in 13 patients (41%). method was conducted using 10-fold cross-validation. The Ocular involvement was detected in 4 patients (12.5%), and predictive accuracy of the PPS was assessed by Harrell tumor cell in CSF was positive in 2 patients (6.2%). An concordance index. elevated LDH serum level was detected in 15 patients The Kaplan–Meier method was used to estimate the (46.8%), and an elevated concentration of CSF protein was survival distribution for each group. A log-rank test was detected in 7 of 12 patients (58.3%) assessed. The samples used to test the differences between the survival groups. The were obtained by stereotactic or open biopsy in 15 patients association of the PPS with OS was evaluated by multivar- (46.8%) and surgical resection in 17 patients (53.1%). The iate analyses with clinical characteristics as other predictors treatments were 3 g/m2 per course for 3 or more cycles of using the Cox proportional hazards regression model. A HD-MTX in 16 cases (50%) and HD-MTX–containing poly- value of P < 0.05 was considered to indicate statistical chemotherapy (cyclophosphamide, pirarubicin, etoposide, significance. vincristine, procarbazine with or without rituximab; ref. 17) in 16 cases (50%). Chemotherapy alone was used in 8 patients (25%) and chemotherapy followed by radiother- Results apy in 24 patients (75%). Radiotherapy was administered to Patient characteristics the whole brain at 30 Gy and local brain at 20 Gy in 12 The baseline characteristics are shown in Table 1. Expres- patients (37.5%), whole brain at 40 Gy in 9 patients sion profiling was conducted on 32 PCNSLs in the test set. (28.1%) and 20 Gy in 3 patients (9.3%). Relapse after The median age of the patients was 64.1 years (range, 44–76 response to the first-line therapy occurred in 19 patients

5674 Clin Cancer Res; 18(20) October 15, 2012 Clinical Cancer Research

Biomarker Study for PCNSL

(relapse rate, 59.3%). The second-line treatment was at the ities and expressions for 6 selected genes (BRCA1, FANCA, physician’s choice. The median OS was 1,626 days in all PPP3R1, RBBP8, ROCK1, and ZNF681). Validation of the patients. Ten patients (31.2%) remained alive (10 with no microarray results was accomplished using qPCR. These 12 evidence of disease) after a median follow-up of 48 months genes were also found to be differentially expressed between (range, 3.8–135.1 months). The causes of death were short-term survivors (survival time, 2.5 years; n ¼ 12) and lymphoma in 16 patients (72.7%), unrelated causes with long-term survivors (survival time, 4 years; n ¼ 9; Sup- no evidence of disease in 3 patients (13.6%), and unknown plementary Table S2). causes with no evidence of disease in 3 patients (13.6%). The patient characteristics in the validation set are similar Survival analysis using the selected 23-gene classifiers to the test set except more patients are involved in deep reveals a prognostic value lesion (Table 1). The patients were monitored for tumor Kaplan–Meier curves were drawn for groups classified by recurrences during the initial and maintenance therapy by clustering analyses based on the gene expressions selected MRI or CT. by the significance analysis of microarrays (SAM; ref. 18) with the false discovery rates (Fig. 1A) and by the random Selection of predictive genes survival forests model (Fig. 1B). The corresponding P values Microarray data have been deposited in Gene Expression by the log-rank test were P ¼ 0.038 for the SAM and P < Omnibus (accession number GSE34771). Twenty-three 0.0001 for the random survival forests model using the 23- genes were selected as the predictors. Table 2 shows a list gene set. These results show that the random survival forests of the genes with their obtained variable importance values. model is more useful than direct use of the gene expressions. Variable importance measures the increase (or decrease) in the prediction error for the random forests model when a Survival analysis using the selected 23-gene classifiers variable is randomly "noise up." That is, if the prediction reveals a prognostic value independent of HD-MTX– error of the model became worse when the effect of one based chemotherapy regimens variable in the model on the prediction was intentionally The 23-gene profile was tested for the prediction of destroyed, this means that the variable is important in outcome in the HD-MTX and HD-MTX–containing poly- the model. The scatter plot in Supplementary Fig. S1 shows chemotherapy groups using Kaplan–Meier curves. The cor- the relationships between the estimated ensemble mortal- responding P values for the log-rank test were P ¼ 0.0001 for

Table 2. Identiﬁcation of survival-related 23 genes

Probe Symbol Description VI 209092_s_at GLOD4 Glyoxalase domain containing 4 0.0368 238962_at ZNF681 Zinc ﬁnger protein 681 0.0255 223779_at AFAP1AS AFAP1 antisense RNA (nonprotein coding) 0.0227 203344_s_at RBBP8 Retinoblastoma binding protein 8 0.0170 201839_s_at EPCAM Epithelial cell adhesion molecule 0.0170 236976_at FANCA Fanconi anemia, complementation group A 0.0113 200886_s_at PGAM1 Phosphoglyceratemutase 1 (brain) 0.0113 213044_at ROCK1 Rho-associated, coiled-coil containing protein kinase 1 0.0085 224874_at POLR1D Polymerase (RNA) I polypeptide D, 16 kDa 0.0085 209146_at SC4MOL Sterol-C4-methyl oxidase-like 0.0085 239233_at CCDC88A Coiled-coil domain containing 88A 0.0085 224850_at ATAD1 ATPase family, AAA domain containing 1 0.0057 236302_at PPM1E Protein phosphatase, Mg2þ/Mn2þ dependent, 1E 0.0057 220176_at NUBPL Nucleotide binding protein-like 0.0057 204531_s_at BRCA1 Breast cancer 1, early onset 0.0028 203560_at GGH g-Glutamyl hydrolase (conjugase, folylpolygammaglutamyl hydrolase) 0.0028 226103_at NEXN Nexilin (F-actin binding protein) 0.0028 217398_x_at GAPDH Glyceraldehyde-3-phosphate dehydrogenase 0.0000 232881_at GNASAS GNAS antisense RNA (nonprotein coding) 0.0028 223721_s_at DNAJC12 DnaJ (Hsp40) homolog, subfamily C, member 12 0.0028 204507_s_at PPP3R1 Protein phosphatase 3, regulatory subunit B, a 0.0057 205339_at STIL SCL/TAL1 interrupting locus 0.0085 233970_s_at TRMT6 tRNAmethyltransferase 6 homolog (S. cerevisiae) 0.0142

Abbreviation: VI, variable importance.

www.aacrjournals.org Clin Cancer Res; 18(20) October 15, 2012 5675

Kawaguchi et al.

A SAM B 23 Gene classifiers

P = 0.0380 P < 0.0001 Percent survival (%) Percent survival

Figure 1. Survival analyses using 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 the selected 23-gene classifiers 0 1,000 2,000 3,000 4,000 0 1,000 2,000 3,000 4,000 reveal a prognostic value. A, Days Kaplan–Meier curves comparing C HD-MTX group D HD-MTX polychemo group groups classified by the clustering analysis based on the gene expressions selected by the SAM. B, comparison of groups classified P = 0.0001 P < 0.0001 by the fitted random survival forests model with the 23-gene model. C, Kaplan–Meier curves comparing groups classified by the fitted random survival forests model with the 23-gene model for patients treated with HD-MTX and (D) HD-MTX–containing polychemotherapy. E, Kaplan– Meier curves comparing groups classified by the Z1 PPS with the 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 23-gene model. F, Kaplan–Meier 0 500 1,000 1,500 2,000 2,500 3,000 0 1,000 2,000 3,000 4,000 curves comparing groups classified by the Z PPS with age, Z Z 2 E 1 score F 2 score KPS, PPP3R1, RBBP8, and BRCA1 IHC.

P < 0.0001 P = 0.0004

Z ≦ Z ≦ 1 1.82 2 3.48

Z 2 ᧺3.48 Z 1 ᧺1.82 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0 1,000 2,000 3,000 4,000 0 1,000 2,000 3,000 4,000

the HD-MTX chemotherapy group (Fig. 1C) and P < 0.0001 0.28 BRCA1 þ 0.26 ATAD1 þ 0.27 GGH þ 0.22 for the HD-MTX–containing polychemotherapy group (Fig. GLOD4 þ 0.15 EPCAM þ 0.14 AFAP1AS þ 0.23 1D). These results show that the random survival forests POLR1D þ 0.1 NEXN þ 0.14 PPM1E þ 0.15 model is useful for predicting survival irrespective of the SC4MOL þ 0.22 NUBPL þ 0.27 CCDC88A þ 0.04 HD-MTX–based chemotherapy regimens. DNAJC12. The Z1-score of the expression value for each individual Identiﬁcation of a PPS associated with survival gene was adapted in this formula. The Z1 scores ranged from The PPS was computed from a linear combination of the 10.0 to 4.43, with a high score associated with a poor 23 genes and calculated for each tumor as follows: Z1 ¼ 0.18 outcome. The optimal cut-off was a Z1 score of 1.82. As ZNF681 þ 0.03 GNASAS þ 0.15 FANCA þ 0.06 expected, the predictor performed well in term of the GAPDH þ 0.28 TRMT6 þ 0.24 PGAM1 þ 0.28 prognosis: the good prognosis group (Z1 1.82) had a PPP3R1 þ 0.23 RBBP8 þ 0.27 ROCK1 þ 0.26 STIL þ median survival time of 2,271 days, whereas the poor

5676 Clin Cancer Res; 18(20) October 15, 2012 Clinical Cancer Research

Biomarker Study for PCNSL

prognosis group (Z1 > 1.82) had a median survival time of gene expression–based method according to Wright and 640 days (P < 0.0001; Fig. 1E). The 10-fold cross-validated c- colleagues (19). Ten cases were classified as GCB and 9 index was 0.856 [95% confidence interval (CI), 0.824– were classified as ABC (Supplementary Fig. S3A). There 0.887; P < 0.0001], indicating a significant predictive was difference in survival between these 2 groups in uni- accuracy. variate analyses (P ¼ 0.026; Supplementary Fig. S3B), and The potential extension of the microarray-based outcome no difference in multivariate analyses (Table 4). We also prediction to the clinical setting was further explored using immunophenotyped the cases in the GCB and ABC sub- IHC detection methods. For practical purposes, we tried to groups by CD10, BCL-6, and MUM1 IHC according to incorporate the IHC data and clinical parameters to com- Camilleri-Bro€et and colleagues (20). Six cases were classi- pute another PPS. The following formula was constructed fied as GCB and 21 were classified as ABC. There was no by the Cox proportional hazards regression model and difference in survival between these 2 groups (P ¼ 0.256; backward selection: data not shown). In our cases, classification by cell-of-origin by microarray was not significantly associated with patient Z2 ¼ 0:04 AGE 0:58 KPS þ 0:22 PPP3R1 survival. þ 1:42 BRCA1 þ 1:11 RBBP8

where KPS was scored as 1 for 70 to 100 and 0 for 10 to 60, High BRCA1 expression is associated with poor and PPP3R1, RBBP8, and BRCA1 were scored as 1 for survival immunohistochemical scores of 1 or more and 0 for immu- BRCA1 mRNA determined by qPCR was found to be nohistochemical score of 0. differentially expressed between short-term and long-term Z The 2 scores ranged from 2.15 to 4.99, with a high score survivors (P ¼ 0.027; Supplementary Table S2). Examples of associated with a poor outcome. The optimal cut-off with BRCA1 IHC are shown in Fig. 2B. The result of double- Z highest value of the log-rank test was a 2 score of 3.48. As immunostaining of BRCA1 and CD79a (B-cell marker) expected, the predictor performed well in term of the showed that BRCA1-positive signals were detected in nucle- Z prognosis: the good prognosis group ( 2 3.48) had a us of CD79a-positive lymphoma cells, which had enlarged median survival time of 2,271 days, whereas the poor nucleus. The staining pattern for BRCA1 was predominantly Z prognosis group ( 2 > 3.48) had a median survival time nuclear in 8 cases, cytoplasmic in 7 cases, and both nuclear P ¼ of 721 days ( 0.0004; Fig. 1F). It should be noted that we and cytoplasmic in 4 cases. There was no significant differ- used different methods to compute the PPS for clinical ence in OS between the nuclear and cytoplasmic patterns characteristics owing to poor significance. (data not shown). However, there was a significant differ- The gene expression predictor is the most significant ence in the PFS between the BRCA1-positive and BRCA1- feature negative groups in both datasets (Fig. 2C and D). The PFS The performance of the gene expression predictor (PPS) was defined as the time from first diagnosis to disease was compared with those of traditional individual features. recurrence or death in univariate analyses. Overexpression of BRCA1 mRNA or protein was strongly associated with As shown in Table 3, Z1 and Z2 were significantly associated with OS in the univariate analyses. Other prognostic scoring poor survival in patients with PCNSL. However, FANCA, systems, such as IELSG (9) or MSKCC (10) prognostic risk PPP3R1, ROCK1, and RBBP8 IHC findings were not signif- group classification were not significant in our series. Table 4 icantly associated with patient survival (Table 3). shows the results for multivariate analyses, in which the Z clinical characteristics were treated as 2 as shown in Table Discussion 4A or selected by the stepwise procedure as shown in Table The reason little progress in molecular analyses of PCNSL 4B. As shown in Table 4, the gene expression predictor Z was 1 has been achieved so far is the very tiny sample amounts significantly associated with OS in the multivariate analyses. obtained for genetic analyses. Although, our study is still It should be noted that BRCA1 IHC was associated with associated with a small number of patients, it is the largest progression-free survival (PFS) in the univariate analyses and series to date and the first study using a gene expression nearly with OS in the multivariate analyses. prognostic classification context in patients with PCNSL. A better understanding of PCNSL biology is crucial to Z2 formula was validated in the independent sample set Because validation of the gene expression signature in improve its prognosis. However, only a few studies have been reported on gene expression profiles of PCNSLs. another independent set is difficult, Z2 score was validated Rubenstein and colleagues (21) compared the gene expres- in the validation set (Table 1). The Z2 scores ranged from 0.97 to 6.36. As expected, there was a significant difference in sion signature of 23 patients with PCNSL with that of 9 patients with nodal large B-cell lymphoma. They showed the OS between the good prognosis group (Z2 3.48) and that individual cases of PCNSL were classified as GCB cell, the poor prognosis group (Z2 > 3.48; P ¼ 0.0281; Fig. 2A). ABC cell, or type III large B-cell lymphoma based on the cell- Classification by cell-of-origin is not associated with of-origin classification described by Alizadeh and collea- survival gues (22). In addition, PCNSLs were distinguished from We classified our cases in germinal center B-cell–like nodal B-cell lymphoma by high expression of regulators of (GCB) and activated B-cell–like (ABC) subgroups using a the unfolded protein response signaling pathway by c-Myc

www.aacrjournals.org Clin Cancer Res; 18(20) October 15, 2012 5677

Kawaguchi et al.

Table 3. Prognostic value of clinical factors Table 3. Prognostic value of clinical factors stratified by OS and PFS in patients with PCNSL stratified by OS and PFS in patients with PCNSL (Cont'd ) Median Median Variable OS, wks P PFS, wks P Median Median Variable OS, wks P PFS, wks P Age, y 0.581 0.777 65 286 178 MSKCC 0.557 0.819 65> 184 73 Age 50 495 248 Gender 0.324 0.596 Age > 50, 191 123 Male 311 100 70 KPS Female 232 123 Age > 50, 232 47 KPS 0.256 0.547 70 > KPS 70 286 123 IELSG 0.781 0.827 70> 232 78 0–1 230 111 No. of lesions 0.419 0.782 2–3 232 113 Single 286 135 3–4 223 145 Multiple 122 78 Z1 <0.0001 <0.0001 Deep lesions 0.704 0.777 1.82< 91 34 Yes 135 68 1.82 324 298 No 286 147 Z2 0.0004 0.0059 LDH serum level 0.646 0.199 3.48< 103 41 216 135 43 3.48 324 248 216> 311 178 NOTE: Polychemo, HD-MTX–containing polychemotherapy. CSF protein level 0.24 0.442 Elevated 191 178 Normal 324 147 and Pim-1. The IL-4 signaling pathway is associated with Operation method 0.26 0.564 tumorigenesis and adverse prognosis in patients with Biopsy 163 73 PCNSL (21). Montesinos-Rongen and colleagues (23) Removal 286 123 reported the gene expression profile of 21 PCNSLs. They Chemotherapy 0.733 0.514 showed that PCNSLs resembled late GCB cells in their gene HD-MTX 136 123 expression pattern, and that PCNSLs were distributed Polychemo 298 123 among the spectrum of systemic DLBCLs. Tun and collea- Immunophenotype 0.026 0.098 gues (24) reported a gene expression comparison between (microarray) 13 PCNSLs and 30 nonCNS DLBCLs. PCNSL was charac- GCB 232 78 terized by significant expression of multiple extracellular ABC 79 34 matrix- and adhesion-related pathways. Sung and collea- Immunophenotype 0.256 0.578 gues (25) evaluated 12 patients with PCNSL by comparative (IHC) genomic hybridization and 7 out of the 12 patients by GCB 363 128 expression profiling. They selected 8 candidate genes in ABC 135 100 which expression changes were associated with copy num- BRCA1 (IHC) 0.168 0.016 ber changes. Positive 96 35 Systemic DLBCLs comprise several diseases that differ in Negative 311 178 responsiveness to chemotherapy (26, 27). The GCB cell– RBBP8 (IHC) 0.232 0.141 like subgroup expressed genes characteristic of normal GCB Positive 135 68 cells and were associated with a good outcome, whereas the Negative 286 147 ABC cell–like subgroup expressed genes characteristic of ROCK1 (IHC) 0.931 0.94 activated B cells and were associated with a poor outcome. Positive 136 123 Gene expression analyses of PCNSLs have largely focused Negative 271 89 on normal lymphocyte development, and the cell-of-origin FANCA (IHC) 0.333 0.197 classification method was not associated with significant Positive 232 123 survival differences in multivariate analyses. Moreover, Negative 122 39 prognostic scoring systems, such as IELSG (9) or MSKCC PPP3R1 (IHC) 0.693 0.401 (10) prognostic risk group classification were not significant Positive 136 68 in our series. Therefore, we developed a novel scoring Negative 311 298 system based on molecular markers. We assessed the relationships between gene expressions (Continued on the following page) and survival time using the random survival forests model,

5678 Clin Cancer Res; 18(20) October 15, 2012 Clinical Cancer Research

Biomarker Study for PCNSL

Table 4. Multivariate analysis: PPS and clinical and therapeutic variables associated with OS

Entire series (n ¼ 32)

Variable Subgroup HR (95% CI) P (A)

Z1 Continuous variable 1.73 (1.08–3.52) 0.017

Z2 Continuous variable 2.49 (0.23–39.4) 0.434 (B) Age Continuous variable 0.99 (0.93–1.05) 0.775 KPS 70, 70> 0.75 (0.28–2.06) 0.573 BRCA1 (IHC) Positive/negative 2.94 (0.98–8.86) 0.052

Z1 Continuous variable 1.45 (1.18–1.91) <0.0001

and its performance provided a better classiﬁcation com- ables (genes) much larger than the number of patients. In pared with the SAM and gene expression subgroups. As this regard, a framework of random forests that overcomes discussed in a study by Cordell (28), the functional form this problem would be necessary in the analysis. Genes were should contain gene-by-gene interaction terms. The ran- selected by applying the RSF-VH algorithm. The advantage dom forests method is classiﬁed into a tree-based method, of this method is that no screening of the genes is necessary. which has an advantage in detecting interactions. It has There are many studies on microarray data using univariate been developed for application to data with several vari- analyses for screening, in which potential genes interacting

A B Z 2 score in validation set BRCA1 IHC

P = 0.0281

ᇫ Z ู3.48 2

Figure 2. Z2 PPS and high BRCA1 expression is associated with poor survival. A, Kaplan–Meier curves (%) survival Percent comparing groups classified by the ᇫZ ᧺3.48 Z2 PPS in the validation set. 2 50 µm B, representative IHC results for 20 µm BRCA1. The result of double- immunostaining of BRCA1 (brown) 0.0 0.2 0.4 0.6 0.8 1.0 0 500 1,000 1,500 and CD79a (B-cell marker, green) Days showed that BRCA1-positive signals were detected in nucleus of CD79a- C D positive lymphoma cells, which have BRCA1 IHC in training set BRCA1 IHC in validation set enlarged nucleus. C, Kaplan–Meier curves comparing groups classified by BRCA1 IHC for PFS in the test set. D, Kaplan–Meier curves comparing P = 0.0166 P = 0.0024 groups classified by BRCA1 IHC for PFS in the validated set. Negative

Negative

Positive Positive 0.0 0.2 0.4 0.6 0.8 1.0 0.0 0.2 0.4 0.6 0.8 1.0 0 1,000 2,000 3,000 4,000 0 500 1,000 1,500

www.aacrjournals.org Clin Cancer Res; 18(20) October 15, 2012 5679

Kawaguchi et al.

with other genes may be dropped from the analyses. In this profiles might not only predict the likelihood of short- regard, the RSF-VH algorithm would be more desirable. term survival, but also yield clues on individual genes Among the selected genes, PPP3R1 is a calmodulin-reg- involved in tumor development, progression, and ulated protein phosphatase, which plays an important role response to therapy. Moreover, the ability to distinguish in signal transduction (29), although there are no reports PCNSLs will enable appropriate therapies to be tailored to about its role in cancer development. BRCA1 seems to specific tumor subtypes. Class prediction models based promote cell survival after DNA damage by preventing on defined molecular profiles allow classification of apoptosis and participating in repair pathways (30). In PCNSLs in a manner that will be better correlated with addition, a role of BRCA1 in the cellular response to clinical outcomes. Therefore, identification of these chemotherapy has been discussed (31, 32). The expression molecular subclasses of PCNSLs could greatly facilitate levels of BRCA1 mRNA or protein predicted survival after prognosis prediction and our ability to develop effective chemotherapy for patients with sporadic ovarian (33), treatment protocols. In conclusion, our profiling results breast (34, 35), prostate (36), and non–small cell lung will help to construct a new classification scheme that cancer (37–40). Low levels of BRCA1 expression resulted better assesses these clinical malignancies. in increased sensitivity to platinum therapy and decreased sensitivity to taxane therapy (33, 35, 37–40). Silencing of Disclosure of Potential Conflicts of Interest the BRCA1 gene by promoter hypermethylation was No potential conflicts of interest were disclosed. reported in sporadic breast and ovarian tumors, especially in the presence of loss of heterogeneity (41). We have Authors' Contributions Conception and design: R. Yamanaka provided evidence that BRCA1 expression may represent Development of methodology: A. Kawaguchi, Y. Komohara, N. Tsuchiya, a predictive biomarker of survival in patients with PCNSL. R. Yamanaka We are trying to further investigate the associations between Acquisition of data (provided animals, acquired and managed patients, provided facilities, etc.): Y. Iwadate, M. Sano, K. Kajiwara, N. Yajima, N. BRCA1 expression and the chemotherapeutic response by Tsuchiya, J. Homma, H. Aoki, T. Kobayashi, Y. Sakai, H. Hondoh, Y. Fujii, R. MTX. Furthermore, as BRCA1 shows promise as a prognos- Yamanaka Analysis and interpretation of data (e.g., statistical analysis, biosta- tic and predictive marker in PCNSL, patients identified as tistics, computational analysis): A. Kawaguchi, Y. Iwadate, T. Kakuma, R. being high expressors could be treated with agents that Yamanaka downregulate BRCA1, thereby sensitizing them to standard Writing, review, and/or revision of the manuscript: A. Kawaguchi, R. Yamanaka therapies. RBBP8 (also known as CtIP) is a BRCA1-inter- Administrative, technical, or material support (i.e., reporting or orga- acting protein (42) and implicated the functional involve- nizing data, constructing databases): A. Kawaguchi, Y. Sakai, Y. Fujii, R. ment in the development of tamoxifen resistance for breast Yamanaka Study supervision: R. Yamanaka cancer (43). Furthermore, work, both in vitro studies and Pathological diagnosis, analyzation, and interpretation of the immu- clinical trials, is needed to assess the correlations between nohistochemical data: Y. Komohara BRCA1-RBBP8 complex expression levels and responses to Carrying out experiments and analyzing data: Y. Sakai potential-targeted therapies. Acknowledgments Because these are retrospective analyses, there are lim- The authors thank Akiyoshi Kakita of Resource Branch for Brain Disease itations and other limitations inherent in a retrospective Research, Brain Research Institute, Niigata University for preparing design. So, these results should be investigated further in specimens. the future. Our PPS may help to identify the patients with PCNSLwhoareunlikelytobecuredbystandardtherapy. Grant Support This work was supported in part by JSPS KAKENHI grant number Our PPS involves a small number of genes, and thus 21700312 to A. Kawaguchi and 20390392 to R. Yamanaka, and by the quantitative reverse transcriptase PCR assays or custom- Collaborative Research Project grant number 2010–2022 to R. Yamanaka of ized DNA microarrays could be developed for clinical the Brain Research Institute, Niigata University. The costs of publication of this article were defrayed in part by the applications. Much more aggressive therapies, such as payment of page charges. This article must therefore be hereby marked high-dose chemotherapy with stem cell implantation advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate (44) or molecular-targeted therapies that specifically tar- this fact. get disabled pathways, might be tailored in those patients Received February 20, 2012; revised July 10, 2012; accepted August 2, with a poor prognosis. In this regard, the expression 2012; published OnlineFirst August 20, 2012.

References 1. Panageas KS, Elkin EB, DeAngelis LM, Ben-Porat L, Abrey LE. Trends primary central nervous system lymphoma in Norway 1989–2003: time in survival from primary central nervous system lymphoma, 1975– trends in a 15-year national survey. Cancer 2007;110:1803–14. 1999: a population-based analysis. Cancer 2005;104:2466–72. 4. Makino K, Nakamura H, Kino T, Takeshima H, Kuratsu J. Rising 2. Olson JE, Janney CA, Rao RD, Cerhan JR, Kurtin PJ, Schiff D, et al. The incidence of primary central nervous system lymphoma in Kumamoto, continuing increase in the incidence of primary central nervous system Japan. Surg Neurol 2006;66:503–6. non-Hodgkin lymphoma: a surveillance, epidemiology, and end results 5. Gavrilovic IT, Hormigo A, Yahalom J, DeAngelis LM, Abrey LE. Long- analysis. Cancer 2002;95:1504–10. term follow-up of high-dose methotrexate-based therapy with and 3. Haldorsen IS, Krossnes BK, Aarseth JH, Scheie D, Johannesen TB, without whole brain irradiation for newly diagnosed primary CNS Mella O, et al. Increasing incidence and continued dismal outcome of lymphoma. J Clin Oncol 2006;24:4570–4.

5680 Clin Cancer Res; 18(20) October 15, 2012 Clinical Cancer Research

Biomarker Study for PCNSL

6. Ferreri AJ, Reni M, Foppoli M, Martelli M, Pangalis GA, Frezzato M, 25. Sung CO, Kim SC, Karnan S, Karube K, Shin HJ, Nam DH, et al. et al. High-dose cytarabine plus high-dose methotrexate versus Genomic profiling combined with gene expression profiling in primary high-dose methotrexate alone in patients with primary CNS central nervous system lymphoma. Blood 2011:117:1291–300. lymphoma: a randomised phase 2 trial.Lancet 2009;374:1 26. Shipp MA, Ross KN, Tamayo P, Weng AP, Kutok JL, Aguiar RC, et al. 512–20. Diffuse large B-cell lymphoma outcome prediction by gene-expres- 7. Chamberlain MC, Johnston SK. High-dose methotrexate and ritux- sion profiling and supervised machine learning. Nat Med 2002;8: imab with deferred radiotherapy for newly diagnosed primary B-cell 68–74. CNS lymphoma. Neuro Oncol 2010;12:736–44. 27. Rosenwald A, Wright G, Chan WC, Connors JM, Campo E, Fisher RI, 8. Thiel E, Korfel A, Martus P, Kanz L, Griesinger F, Rauch M, et al. High- et al. Lymphoma/Leukemia Molecular Profiling Project. The use of dose methotrexate with or without whole brain radiotherapy for pri- molecular profiling to predict survival after chemotherapy for diffuse mary CNS lymphoma (G-PCNSL-SG-1): a phase 3, randomised, non- large-B-cell lymphoma. N Engl J Med 2002;346:1937–47. inferiority trial.Lancet Oncol 2010;11:1036–47. 28. Cordell HJ. Detecting gene-gene interactions that underlie human 9. Ferreri AJ, Blay JY, Reni M, Pasini F, Spina M, Ambrosetti A, et al. diseases. Nat Rev Genet 2009;10:392–404. Prognostic scoring system for primary CNS lymphomas: the Interna- 29. Wang MG, Yi H, Guerini D, Klee CB, McBride OW. Calcineurin A alpha tional Extranodal Lymphoma Study Group experience. J Clin Oncol (PPP3CA), calcineurin A beta (PPP3CB) and calcineurin B (PPP3R1) 2003;21:266–72. are located on human chromosomes 4, 10q21–>q22 and 2p16–>p15 10. Abrey LE, Ben-Porat L, Panageas KS, Yahalom J, Berkey B, Curran W, respectively. Cytogenet Cell Genet 1996;72:236–41. et al. Primary central nervous system lymphoma: the Memorial Sloan- 30. Kennedy RD, Quinn JE, Johnston PG, Harkin DP. BRCA1: mechan- Kettering Cancer Center prognostic model. J Clin Oncol 2006;24: isms of inactivation and implications for management of patients. 5711–5. Lancet 2002;360:1007–14. 11. Shenkier TN, Voss N, Chhanabhai M, Fairey R, Gascoyne RD, Hoskins 31. Kennedy RD, Quinn JE, Mullan PB, Johnston PG, Harkin DP. The role P, et al. The treatment of primary central nervous system lymphoma in of BRCA1 in the cellular response to chemotherapy.J Natl Cancer Inst 122 immunocompetent patients: a population-based study of suc- 2004;96:1659–68. cessively treated cohorts from the British Colombia Cancer Agency. 32. Clark-Knowles KV, O'Brien AM, Weberpals JI. BRCA1 as a therapeutic Cancer 2005;103:1008–17. target in sporadic epithelial ovarian cancer. J Oncol 2010;2010: 12. Livak KJ, Schmittgen TD. Analysis of relative gene expression data 891059. using real-time quantitative PCR and the 2(Delta Delta C(T)) method. 33. Quinn JE, James CR, Stewart GE, Mulligan JM, White P, Chang GK , Methods 2001;25:402–8. et al. BRCA1 mRNA expression levels predict for overall survival in 13. R Development Core Team. R. A language and environment for ovarian cancer after chemotherapy. Clin Cancer Res 2007;13:7413–20. statistical computing. Vienna, Austria: R Foundation for Statistical 34. Rakha EA, El-Sheikh SE, Kandil MA, El-Sayed ME, Green AR, Ellis IO. Computing; 2009. Available from: www.R-project.org. Expression of BRCA1 protein in breast cancer and its prognostic 14. Gentleman R, Carey V, Bates D, Bolstad B, Dettling M, Dudoit S, et al. significance. Hum Pathol 2008;39:857–65. Bioconductor: open software development for computational biology 35. Margeli M, Cirauqui B, Castella E, Tapia G, Costa C, Gimenez-Capitan and bioinformatics. Genome Biol 2004;5:R80. A, et al. The prognostic value of BRCA1 mRNA expression levels 15. Ishwaran H, Kogalur UB, Gorodeski EZ, Minn AJ, Lauer MS. High- following neoadjuvant chemotherapy in breast cancer. PLoS ONE dimensional variable selection for survival data. J Am Stat Assoc 2010;5:e9499. 2010;105:205–17. 36. Forentino M, Judson G, Penney K, Flavin R, Stark J, Fiore C, et al. 16. Youden WJ. Index for rating diagnostic tests. Cancer 1950;3:32–5. Immunohistochemical expression of BRCA1 and lethal prostate can- 17. Yamanaka R, Shinbo Y, Sano M, Homma J, Tsuchiya N, Yajima N, et al. cer. Cancer Res 2010;70:3136–9. Salvage therapy and late neurotoxicity in patients with recurrent 37. Taron M, Rosell R, Felip E, Mendez P, Souglakos J, Ronco MS, et al. primary CNS lymphoma treated with a modified ProMACE-MOPP BRCA1 mRNA expression levels as an indicator of chemoresistance in hybrid regimen.Leuk Lymphoma 2007;48:1119–26. lung cancer. Hum Mol Genet 2004;13:2443–9. 18. Tusher VG, Tibshirani R, Chu G. Significance analysis of microarrays 38. Rosell R, Skrzypski M, Jassem E, Taron M, Bartolucci R, Sanchez JJ, applied to the ionizing radiation response. Proc Natl Acad Sci U S A et al. BRCA1: a novel prognostic factor in resected non–small-cell lung 2001;98:5116–21. cancer. PLoS ONE 2007;2:e1129. 19. Wright G, Tan B, Rosenwald A, Hurt EH, Wiestner A, Staudt LM. A gene 39. Boukovinas I, Papadaki C, Mendez P, Taron M, Mavroudis D, Kout- expression-based method to diagnose clinically distinct subgroups of sopoulos A, et al. Tumor BRCA1, RRM1 and RRM2 mRNA expression diffuse large B cell lymphoma. Proc Natl Acad Sci U S A 2003;100: levels and clinical response to first-line gemcitabine plus docetaxel in 9991–6. non-small-cell lung cancer patients. PLoS ONE 2008;3:e3695. 20. Camilleri-Broet€ S, Criniere E, Broet€ P, Delwail V, Mokhtari K, Moreau A, 40. Rosell R, Perez-Roca L, Sanchez JJ, Cobo M, Moran T, Chaib I, et al. et al. A uniform activated B-cell-like immunophenotype might explain Customized treatment in non–small-cell lung cancer based on EGFR the poor prognosis of primary central nervous system lymphomas: mutations and BRCA1 mRNA expression. PLoS ONE 2009;4:e5133. analysis of 83 cases. Blood 2006;107:190–6. 41. Esteller M, Silva JM, Dominguez G, Bonilla F, Matias-Guiu X, Lerma E, 21. Rubenstein JL, Fridlyand J, Shen A, Aldape K, Ginzinger D, Batchelor et al. Promoter hypermethylation and BRCA1 inactivation in sporadic T, et al. Gene expression and angiotropism in primary CNS lymphoma. breast and ovarian tumors.J Natl Cancer Inst 2000;92:564–9. Blood 2006;107:3716–23. 42. Wang B, Matsuoka S, Ballif BA, Zhang D, Smogorzewska A, Gygi SP, 22. Alizadeh AA, Eisen MB, Davis RE, Ma C, Lossos IS, Rosenwald A, et al. et al. Abraxas and RAP80 form a BRCA1 protein complex required for Distinct types of diffuse large B-cell lymphoma identified by gene the DNA damage response. Science 2007;316:1194–8. expression profiling. Nature 2000;403:503–11. 43. Wu M, Soler DR, Abba MC, Nunez MI, Baer R, Hatzis C, et al. CtIP 23. Montesinos-Rongen M, Brunn A, Bentink S, Basso K, Lim WK, Klapper silencing as a novel mechanism of tamoxifen resistance in breast W, et al. Gene expression profiling suggests primary central nervous cancer. Mol Cancer Res 2007;5:1285–95. system lymphomas to be derived from a late germinal center B cell. 44. Soussain C, Hoang-Xuan K, Taillandier L, Fourme E, Choquet S, Witz Leukemia 2008;22:400–5. F, et al. Intensive chemotherapy followed by hematopoietic stem-cell 24. Tun HW, Personett D, Baskerville KA, Menke DM, Jaeckle KA, Kreinest rescue for refractory and recurrent primary CNS and intraocular lym- P, et al. Pathway analysis of primary central nervous system lympho- phoma: Societ e Francaise¸ de Greffe de Moelle€ Osseuse-Therapie ma. Blood 2008;111:3200–10. Cellulaire. J Clin Oncol 2009;26:2512–8.

www.aacrjournals.org Clin Cancer Res; 18(20) October 15, 2012 5681

Gene Expression Signature−Based Prognostic Risk Score in Patients with Primary Central Nervous System Lymphoma

Atsushi Kawaguchi, Yasuo Iwadate, Yoshihiro Komohara, et al.

Clin Cancer Res 2012;18:5672-5681. Published OnlineFirst August 20, 2012.

Updated version Access the most recent version of this article at: doi:10.1158/1078-0432.CCR-12-0596

Supplementary Access the most recent supplemental material at: Material http://clincancerres.aacrjournals.org/content/suppl/2012/08/20/1078-0432.CCR-12-0596.DC1

Cited articles This article cites 42 articles, 13 of which you can access for free at: http://clincancerres.aacrjournals.org/content/18/20/5672.full#ref-list-1

Citing articles This article has been cited by 2 HighWire-hosted articles. Access the articles at: http://clincancerres.aacrjournals.org/content/18/20/5672.full#related-urls

E-mail alerts Sign up to receive free email-alerts related to this article or journal.

Reprints and To order reprints of this article or to subscribe to the journal, contact the AACR Publications Department at Subscriptions [email protected].

Permissions To request permission to re-use all or part of this article, use this link http://clincancerres.aacrjournals.org/content/18/20/5672. Click on "Request Permissions" which will take you to the Copyright Clearance Center's (CCC) Rightslink site.