Gene Expression Profiles of Central Nervous System Lymphoma Predict Poor Survival in Patients with Diffuse Large B-cell Lymphoma
Yuki Kagoya,Yasuhito Nannya,Fumihiko Nakamura,Mineo Kurokawa
DOI: https://doi.org/10.1111/bjh.12902
2014-01-01
British Journal of Haematology
Abstract:The overall prognosis of B-cell non-Hodgkin lymphoma has substantially improved since the introduction of rituximab. In the treatment of the patients with diffuse large B-cell lymphoma (DLBCL), the addition of rituximab to CHOP (cyclophosphamide, doxorubicin, vincristine, prednisone) therapy (R-CHOP) can achieve much better responses than CHOP alone (Coiffier et al, 2002; Pfreundschuh et al, 2006). In contrast with the effective treatment outcome of systemic lesions, central nervous system (CNS) involvement in DLBCL still poses an insufficient prognosis. One reason is that penetration of most of the chemotherapeutic agents, including rituximab, into CNS parenchyma is limited due to the blood-brain barrier. In addition to this problem of drug delivery, there is a possibility that lymphoma cells invading the CNS possess intrinsic resistance to chemotherapy. Recent studies have demonstrated that DLBCL can be divided into two major subgroups according to gene expression profile: germinal centre B-cell-like (GCB) and activated B-cell-like (ABC) subtype; the ABC group has a significantly poorer overall survival (OS) than the GCB group (Alizadeh et al, 2000). We hypothesized that primary CNS lymphoma (PCNSL) also has a characteristic gene expression profile, and investigated if the expression signature could predict prognosis in nodal DLBCL. To obtain a specific gene expression profile for PCNSL, we thoroughly investigated the genes whose expression levels are significantly increased or decreased in PCNSL compared with those in nodal lymphoma, extranodal lymphoma and normal lymph node using published microarray data deposited in GSE11392 and GSE25297 (Tun et al, 2008; Sung et al, 2011) of the Gene Expression Omnibus (GEO) database. A set of CEL files were downloaded from the GEO and normalized using the justRMA function from the Affy package of Bioconductor. The expression values of individual genes were compared between groups using RankProd analysis. Fifteen genes were selected as being significantly elevated or decreased in PCNSL compared with nodal lymphoma, extranodal lymphoma and normal lymph node, as determined by a false discovery rate <0·01 in all the expression data: 13 genes were highly expressed and two were expressed at low levels in PCNSL. Based on each expression value, we calculated a PCNSL score in each sample by adding the Z scores of the expression levels of individual genes elevated in PCNSL and subtracting the Z scores of the expression measurements of genes that are decreased in PCNSL (Fig. 1A). In order to confirm that our scoring system clearly reflected the gene expression signature of PCNSL, we calculated the PCNSL score in another set of microarray data (GSE10524) comparing the gene expression profiles of PCNSL, nodal lymphoma and testicular lymphoma (Booman et al, 2008). As expected, most of the PCNSL samples had much higher scores than those from nodal lymphoma (P < 0·01). Interestingly, testicular lymphoma cells also had a relatively higher score compared with the nodal lymphoma cells (P < 0·01), suggesting that they have similar biological characteristics to PCNSL. We then investigated the impact of the PCNSL gene expression score on the overall prognosis in nodal lymphoma. OS in 203 patients with DLBCL or primary mediastinal B-cell lymphoma (PMBL) was compared to the groups with high or low PCNSL score using the published data (GSE11318; Lenz et al, 2008). We calculated the PCNSL score in each sample, assigned the upper half into subjects with a high PCNSL score, and the lower half into ones with low PCNSL score, and compared the OS data between the groups. Surprisingly, patients with high PCNSL scores had significantly worse prognosis compared with those with low scores (P < 0·01) (Fig. 1C). We also examined the correlation between the GCB/ABC subtype and our PCNSL score. Lymphoma cells with ABC subtype had significantly higher PCNSL score than those with GCB subtype (Fig. 1D). Interestingly, PMBL scores were as high as those of ABC-DLBCL. Although there was a positive correlation between ABC subtype and high PCNSL score, a high PCNSL score was significantly associated with worse prognosis after adjustment for the International Prognostic Index (IPI) and GCB/ABC subtype classification (Table 1; hazard ratio 1·82, 95% confidence interval 1·39–2·25, P < 0·01). Furthermore, when only patients with ABC-DLBCL were analysed for OS according to the PCNSL score, patients with a high score still had a tendency to worse OS (P = 0·06; Fig. 1E). These results show that PCNSL has a characteristic gene expression profile, represented by selected genes, and that the gene expression signature was significantly associated with poor prognosis in patients with DLBCL or PMBL. Among DLBCL, samples with ABC subtype had higher PCNSL scores compared to those with GCB subtype, which is consistent with a previous study showing that most of PCNSL cases are of the ABC subtype (Camilleri-Broët et al, 2006). We showed, however, that high PCNSL gene expression signature was significantly associated with worse OS, independent of IPI and ABC subtype. Therefore, we consider that there are gene expression profiles uniquely related to PCNSL, and the signature is associated with overall prognosis in nodal lymphomas. Another interesting finding in our study is that testicular lymphoma had a relatively high PCNSL score compared with nodal lymphoma. Given that testis involvement is one of the most striking risk factors for secondary CNS involvement (Villa et al, 2010), there is a possibility that high PCNSL score in lymphoma cells at diagnosis reflects an increased risk for secondary involvement of the CNS, which might be useful for designing preventive treatment in high-risk patients. Worse OS in patients with high PCNSL scores is also consistent with this possibility, considering that CNS relapse inevitably leads to poor prognosis (Kridel & Dietrich, 2011). Further studies are required to verify the influence of initial gene expression profiles of DLBCL on the risk for subsequent CNS relapse, which might contribute to the stratification of treatment regimens according to patient risk. In conclusion, we identified selected genes associated with the PCNSL gene expression signature. High PCNSL score was significantly associated with poor prognosis in patients with DLBCL and PMBL, and might be useful for stratification of treatment protocols including the prophylaxis for CNS relapse. YK designed the study, performed data analysis and wrote the first draft of the article. YN, FN, and MK edited the article and supervised the data analysis. All authors are responsible for the data interpretation and approved the final version of the article. The authors have declared that no conflict of interest exists.