New genetic prognostic biomarkers in primary central nervous system lymphoma (PCNSL)

Abstract Background PCNSL is a rare extranodal NHL with poor prognosis. Tumorigenesis has been associated with hyperactivation of BCR downstream and NFkB pathways. We studied the prognosis of the relative expression profile of target genes of NFkB pathway (MYC, BCL2), the essential transcriptional regulator in hematopoiesis LMO2, the checkpoint regulation pathway MGMT, the transcription factor POU2F1, the immune checkpoint gene PDCD1, and the proto‐oncogene and transcriptional repressor gene BCL6 and its proteins in PCNSL. Methods This study is a retrospective cohort study; 35 immunocompetent PCNSL‐DLBCL patients had their gene expression (RT‐qPCR) normalized to internal control gene GUSB. Results Median patient age was 62 years, median OS was 42.6 months (95% CI: 26.6–58.6), PFS was 41 months (95% CI: 19.7–62.4), and DFS was 59.2 months (95% CI 31.9–86.6). A moderate correlation was found between the gene/protein expressions of MYC (kappa = 0.596, p = .022) and of BCL2 (kappa = 0.426, p = .042). Relative gene expression of MYC ≥ 0.201 (HR 6.117; p = .003) was associated with worse 5‐year OS. Relative gene expression of MYC ≥ 0.201 (HR 3.96; p = .016) and MGMT ≥ 0.335 (HR 3.749; p = .056) was associated with worse PFS. Age > 60 years and IELSG score moderate/high were also associated with worse prognosis. Conclusions Overexpression of MYC and overexpression of MGMT were prognostic markers associated with unfavorable clinical outcomes in PCNSL.

Despite its rarity (incidence of 0.7/100.000 persons per year), the rate has been increasing by 1.6% per year in the last decade, mainly in elderly people (Shiels et al., 2016).
It is described that enrichment of mutations involving the B-cell receptor signaling pathway and hyperactivation of the NF-kappaB prosurvival signaling pathway (NFkB) are associated with the pathogenesis of PCNSL; however, the rarity of this disease and the paucity of histopathological material available area challenge for molecular studies (Braggio et al., 2015;Rubenstein, 2017).
Some studies have suggested the gene expression of MYC, BCL2, LMO2, MGMT, POU2F1, PDCD1, and BCL6 as prognostic biomarkers in PCNLS. MYC is a proto-oncogene, and its protein is a nuclear transcription factor involved in cell cycle progression and apoptosis, being associated with worse prognosis in patients with systemic DLBCL (Kawamoto et al., 2016;Perry et al., 2014). Its expression is upregulated in PCNSL, which may lead to lymphomagenesis (Fischer et al., 2011;Grommes et al., 2017). Copy number alterations and translocations at chromosome 9p24 have been described, involving the genes coding for programmed death ligand 1 (PD-L1) and PD-L2, suggesting that immune escape may also be involved in the pathogenesis of PCNSL, with potential benefit of programmed cell death 1 (PD-1) blockade therapy (Chapuy et al., 2016). MGMT, a tumor suppressor gene that has been associated with PCNSL , codifies a protein that repairs DNA, and may be silenced by methylation of the promotor region. Overexpression of MGMT is associated with worse prognosis, and methylation of its promotor region was associated with better overall survival among patients treated with high-dose chemotherapy, and with a better tumor response in patients treated with temozolomide (Kurzwelly et al., 2010).
PCNSL often presents translocation involving immunoglobulin (Ig)-related genes, including BCL6 (Basso & Dalla-Favera, 2012), leading to constitutive activity and to tumorigenesis (Cattoretti et al., 2005). BCL2 is a proto-oncogene regulating cell cycle and apoptosis, and its expression is associated with lymphomagenesis (Vaux et al., 1988) and is a biomarker of worse prognosis in DLBCL (Kawamoto et al., 2016;Perry et al., 2014). LMO2 forms a transcriptional complex that regulates gene expression in DLBCL and is usually overregulated in NHL and is associated with improved prognosis in both DLBCL and PCNSL, being overexpressed in 52% of PCNSL (Cubedo et al., 2012;Lossos et al., 2014). POU2F1 is usually overexpressed in NHL, which may lead to higher susceptibility to therapies as irinotecan, paclitaxel, and mutations on POU2F1 may lead to tumor resistance against chemotherapies (Gupta et al., 2012).
The increasing incidence of PCNSL, mainly in elderly population, who frequently presents comorbidities and are more vulnerable to oncological treatments (Abrey et al., 1998;Kasenda et al., 2015), the aggressiveness of this disease, the current toxicity profile of the therapies, and the heterogeneity in patient response to therapy and outcomes (Abrey et al., 2005) demand further research on biomarker identification that could help to identify potential targets for new therapies and provide prognostic information.

| Study population
The Local Institutional Review Board in accordance with the Declaration of Helsinki approved this study (IRB No. 1.266.854, approval date: 7 October 2015 Baseline clinical, laboratory, and disease features were retrieved from medical records. Cranial magnetic resonance imaging (MRI), bone marrow biopsy, and neck, chest, abdomen, and pelvis computerized tomography (CT) data were also collected. International Extranodal Lymphoma Study Group Prognostic Score (IELSG) was calculated as described elsewhere (Ferreri et al., 2003).
Patients were treated with high-dose methotrexate (HDMTX 3.5-5 g/m 2 IV, five cycles each 15 days, 3 hr infusion) combined or not with other chemotherapeutical agents as induction treatment.
Consolidation therapy was prescribed for patients who achieved complete response (CR) after induction therapy, consisting of wholebrain radiotherapy (WBRT) or up to 2 cycles of cytarabine (2 g/ m 2 IV, on D1 and D2). Patients with poor clinical conditions were treated with WBRT (30-40Gy, up to 20 fractions) or best support care. Tumor response was evaluated according to the International PCNSL Collaborative Group (IPCG), and active follow-up was performed (Abrey et al., 2005).

| Immunohistochemistry
Two hematopathology experts reviewed the tumor histology according to the WHO criteria (Swerdlow et al., 2017). Four-millimeter sections from formalin-fixed paraffin-embedded (FFPE) tumor samples obtained at diagnosis were stained with hematoxylin-eosin (HE), as previous described (Grizzle el al., 1998), followed by an initial immunohistochemistry (IHC) panel with monoclonal antibodies (MoAbs) CD3 and CD20. If the initial analysis was suggestive of PCNSL histopathology diffuse large B-cell lymphoma, the IHC panel was expanded (Swerdlow et al., 2017).
Briefly, antigen retrieval was performed using citrate buffer in a Pascal pressure cooker set to a temperature of 125°C and a pressure of 18 psi (Celerus RIPTIDE; Celerus Diagnostics). Endogenous peroxidase was blocked, and the samples were incubated overnight in a solution containing monoclonal antibodies for MYC (clone Y69; Abcam) at a dilution of 1:100 in pH 9.0, BCL2 (clone 124, Cell Marque) at a dilution of 1:100 in pH 9.0, LMO2 (clone CP51; Cell Marque) at a dilution of 1:50 in pH 9.0, MGMT (clone MT3.1; Abcam) at a dilution of 1:400 in pH 9.0, POU2F1 (clone AB151; Abcam) at a dilution of 1:200 in pH 9.0, and PDCD1 (clone NAT 105; Abcam, USA) at a dilution of 1:100 at pH 9.0. Samples were incubated with Novolink Polymer (Leica Biosystems Newcastle) for 30 min and then incubated in the chromogenic substrate diaminobenzidine (DAB, from DBS) and counterstained with hematoxylin. Slides were mounted using a synthetic resin (Entellan; Merck). Due to the lack of enough histopathological material, MoAb for BCL6 was not performed. The positive score was determined as the percentage of positive tumor cells using a semiquantitative method in 25% increments under an optical microscope (40X objective). Two distinct observers analyzed the slides.

| Molecular biology
Total RNA was extracted from FFPE tissues (5μm-thick) using the Qiagen FFPE RNeasy Kit (Valencia) with a modified deparaffinization step (Belder et al., 2016). Briefly, sections were deparaffinized by two repeated incubations in xylene for 10 min, followed by two Real-Time PCR Experiment (MIQE) guidelines (Bustin et al., 2009) and the determined PCR efficiency for each assay using FFPE for GUSB, and all tested genes (all 94.1%-103.5%).

| Statistical analysis
Descriptive statistics were calculated to summarize clinical and molecular characteristics. Gene expression and protein expression by IHC were dichotomized into low and high expression levels using the receiver operating characteristic (ROC) analysis (Choi, 1998).
Optimal cutoff points for relative gene expression and immunohistochemistry were determined based on the best balance of sensitivity and specificity along with larger increases in LRs, which best differentiate clinical outcomes. The univariate analysis to assess the association among categorical variables was performed using the Fisher exact test. A Cox univariate analysis was performed to estimate the association between categorical variables and survival curves (Fischer, 1922). Due to small sample, multivariate analysis was not performed.
Overall Survival (OS), progression-free survival (PFS), and disease-free survival (DFS) were determined using the Kaplan-Meier method. The log-rank test was used to compare survival curves and to verify the association between categorical variables and survival curves. A correlation test between gene expression and IHC was performed, using the Youden index and Kappa coefficient (Youden, 1950). All analyses were performed using Statistical Package for the Social Sciences (SPSS) version 11.0. p values ≤ .05 were considered statistically significant.

| RE SULTS
Fifty-one patients were initially screened and retrospectively evaluated, and 35 patients (68.6%) were enrolled; 16 patients were excluded due to unavailability of paraffin block, medical records, or after histology revision. Patient's characteristics are shown in Table 1. Median follow-up for all patients was 41 months (range, 1-172). Twenty-six patients (74.2%) were dead at the cutoff date, with a median interval between diagnosis and death of 26.6 months (range, 1-92). The main cause of death described was progression disease (PD) in 21 patients (80.7%).
The median OS was 42.6 months (95% CI: 26.6-58.6). The OS rates at 24, 48, and 60 months were 65.5%, 43.8%, and 40.4%, respectively. The median PFS was 41 months (95% CI: 19.7-62.4), and the median DFS was 59.2 months (95% CI 31.9-86.6). Figure 1 shows survival curves. Age higher than 60 years, IELSG score ≥ 2 (moderate/high), ECOG ≥ 3, treatment without WBRT, and refractory disease were statistically significant predictors of worse survival in univariate analysis (Table 2).  respectively. Subjects with absent values of gene expression had the qRT-PCR repeated and confirmed. Comparison between the variables in this subgroup and the total population did not show any clinically significant difference; therefore, the subgroup was representative of the total population.  (Figure 3) in univariate analysis.

| D ISCUSS I ON
Despite its small sample, this study is aligned with most molecular studies with PCNSL, which usually are retrospective cohort studies, with populations usually smaller than 50 patients (Braggio et al., 2015;Mrugala et al., 2009;Rubenstein, 2017). Similar to other studies, mean age of diagnosis was 62 years and approximately 30% of them were older than 70 years. The increasing incidence of PCNSL in elderly population has an impact on the choice of treatment and its tolerability (Siegal & Bairey, 2019). In this trial, despite majority of patients had ECOG PS ≥ 2 and age > 60 years, 94.2% received any kind of oncological therapy with curative intent. Ferreri and al. described that 98% of patients with ECOG PS ≥ 2 were able to receive at least 1 line of therapy (Ferreri et al., 2003). Age > 60 years and ECOG PS ≥ 2 were the first poor prognostic biomarkers identified in PCNSL (Nelson et al., 1992;Schultz et al., 1996) and were also associated with worse prognosis in this cohort.  (Song, 2019). In our study, WBRT was associated with better outcomes (OS, DFS, and PFS); however, due to the small sample, WBRT as consolidation therapy and as monotherapy were analyzed together; therefore, the benefit of WBRT should be interpreted with caution. Also, as the life expectancy of this population increases, bigger attention should be paid to long term toxicities, and WBRT has been associated with a decline in motor functions and learning in 50% of patients up to 2 years after therapy (Citterio et al., 2017), and it was not the focus of this study. Anti-CD20 therapies were not prescribed for patients in this cohort, neither bone marrow transplantation.
There was no gene amplification in between 34.3% and 40% of the samples, which may have occurred due to the long-term storage of the FFPE and the small amount of tissue (Abrahamsen et al., 2003;von Smolinski et al., 2005). Nonetheless, there was no difference statistically significant between the total samples and the amplified samples concerning the variables analyzed. overexpressed, leading to worse prognosis (Mounier et al., 2003).
In pre-rituximab era, BCL2 expression was considered a poor prognostic factor in NHL; however, the overexpression of BCL2 in the post-rituximab era in DLBCL has shown no prognostic value (Mounier et al., 2003). In this trial, no patient was treated with rituximab.
There was a moderate correlation between gene expression and protein expression for MYC and BCL2 (kappa between 0.41 and 0.60). Usually, the transcriptome is not directly proportional to the protein level, being insufficient to explain the correlation between the genotype and its phenotype . Possible reasons include the lack of protein stability, transcription and translational rate, mRNA degradation, and post-translational and posttranscriptional regulations (Bujak et al., 2015;Liu et al., 2016).
MGMT and its protein are usually studied in tumorigenesis due to its potential prognostic impact and as predictor factor to alkylating agents. Its protein repairs mutagenic DNA lesion preventing mismatch and errors during DNA replication and transcription (Cai et al., 2019;Toffolatti et al., 2014). Hypermethylation of the CpG islands in the promotor region of MGMT leads to transcriptional silencing of the DNA repair enzyme O (6)-methylguanine-DNA methyltransferase, being reported in 37% of PCNSL (Zheng et al., 2017).
Methylation has been correlated with a better overall response in patients with multiform glioblastoma and also in PCNSL (Adachi et al., 2012;Thomas et al., 2013;Zheng et al., 2017), therefore is a potential for targeted therapies. In this study, median gene expression was zero and immunohistochemistry was negative for all patients. Methylation of its promotor was not studied; however, others. Some other studies did not confirm its prognostic value (Maddox et al., 2012;Vazquez-Arreguin et al., 2018). POU2F1 is a proto-oncogene, operating target genes associated with proliferation, immune modulation, oxidative, cytotoxic stress resistance, and metabolic regulation, among others. Further data also show that POU2F1 is involved in the high-affinity transport of anthracyclines, and defects on OCT1 could potentially contribute to resistance of cancer cells to some chemotherapeutical agents. In a retrospective study with 77 DLBCL, POU2F1 lower expression was associated with better OS and PFS (Gouveia et al., 2020). In this trial, POU2F1 was not associated with the clinical outcomes studied.
Immune escape mechanisms have also been studied in PCNSL.
PDCD1 is an immune checkpoint and protects against autoimmune responses, and the interaction of PD-L1 on cancer cells with PD1 on T cells leads the cancer cells to escape from immune system (Menter et al., 2016). PD-L1 has been reported in 53%-60% of tumorinfiltrating lymphocytes (TIL) and 37% in tumoral cells in PCSNL (Berghoff et al., 2014), suggesting that immune escape may have a role in PCNSL tumorigenesis. Its prognostic value has been reported by Cho et al, who described that overexpression of PDCD1 protein, was associated with worse OS and DFS, however with different cutoff values used from this study (≥70 cells/HPF) (Cho et al., 2017). In this trial, median gene expression of PDCD1 was zero and the majority of patients (85%) did not present PDCD1 immunoexpression in tumor cell or in the TIL.
In conclusion, overexpression of genes MYC and MGMT was associated with worse outcomes in this study, implying these genes as potential biomarkers related to poor prognosis in PCNSL. Such findings need to be confirmed in future trials with a larger casuistic.

ACK N OWLED G M ENTS
We would like to thank Hospital das Clínicas da Faculdade de Medicina da Universidade de São Paulo, Instituto do Câncer do Estado de São Paulo, for permanent support and interest in our studies.

CO N FLI C T S O F I NTE R E S T
The authors declare no conflict of interest.

PEER R E V I E W
The peer review history for this article is available at https://publo ns.com/publo n/10.1002/brb3.2061.

DATA AVA I L A B I L I T Y S TAT E M E N T
The data that support the findings of this study are available from the corresponding author upon reasonable request.