Introduction

Diffuse large B-cell lymphoma (DLBCL) is the most prevalent type of non-Hodgkin B-cell lymphoma (B-NHL), representing roughly 30–40% of all cases worldwide. Patients commonly present with fast growing tumors which can involve single or multiple, nodal or extranodal sites [1]. DLBCL can virtually develop in any primary tissue site from two major cellular subtypes: activated B cell-like (ABC) or germinal center B cell-like (GCB). The respective subtypes have different mechanisms of development, genetic alterations, and treatment response; with the ABC subtype showing an inferior prognosis [1, 2].

Primary central nervous system lymphoma (PCNSL) is a rare subtype of DLBCL that arises and is confined to the central nervous system (CNS) [3]. This non-Hodgkin aggressive B-cell lymphoma is distinguished from extra-cerebral DLBCL by its poorer prognosis. PCNSL can occur in the setting of immunosuppression (HIV/AIDS, post-transplant) or in immunocompetent individuals [4,5,6]. While treatment response rates are high, relapses are frequent and prognosis after recurrence is poor with 5-year survival rates ranging from 22 to 40% [7, 8]. The genomic alterations (GAs) underlying PCNSL have not been comprehensively studied.

Single nucleotide mutations in various genes, including MYD88, CD79b, PIM1, and BTG2, have been reported as the most prevalent genetic alterations in PCNSL [9,10,11]. Among these mutations, an MYD88 substitution mutation at c.794T > C resulting in a replacement of leucine 265 by proline (L265P) is the most common mutation in PCNSL. Myeloid differentiation factor 88 (MYD88) is an adaptor molecule in the Toll-like receptor pathway that mediates interleukin-1 receptor signaling [12]. Similar to several other common mutations in PCNSL, MYD88 mutations lead to constitutive activation of the nuclear factor NF-κB signaling pathway [13, 14]. Translocation of NF-κB into the nucleus subsequently initiates activation of other target genes [13, 14].

From a molecular perspective, GAs are of great interest as they can serve as diagnostic biomarkers or targets for personalized therapies. Despite considerable progress in the understanding of CNS lymphomas, the majority of existing molecular data is derived from locus specific approaches targeting single candidate genes for point mutations, like MYD88 [15]. Over the past decade, the development and affordability of next generation sequencing (NGS) has facilitated several studies identifying GAs in CNS lymphomas through targeted and whole-exome sequencing [15,16,17,18]. Nonetheless, the rarity of the disease and the restricted availability of affected brain tissue hinder the study of molecular and GAs in PCNSL. Therefore, our understanding of this disease remains limited [19]. In addition, discrimination of PCNSL and secondary CNS lymphomas can be very challenging by conventional microscopic examination and magnetic resonance imaging (MRI) alone [20]. Better understanding of molecular alterations in PCNSL can be of clinical utility by facilitating the distinction of PCNSL from secondary CNS lymphoma. To increase our understanding of GAs in PCNSL, we retrospectively investigated the results of a comprehensive NGS assay in a series of 15 PCNSL.

Methods

Patients and tumor samples

This retrospective study was approved by the institutional review board of the University of Texas Health Science Center at Houston and Memorial Hermann Hospital, Houston, TX. From January 2010 to December 2017, 50 consecutive patients diagnosed with PCNSL were identified in the clinical records of the University of Texas Health Science Center at Houston and Memorial Hermann Hospital, Houston TX. The results of molecular testing were available for 15 patients.

All 15 tumor samples were examined by H&E (Supplementary Fig. S1) and immunohistochemistry, and confirmed as DLBCLs. Two patients had a history of immunosuppression (HIV positive). The patients’ ages ranged from 22 to 80 years, average age of 58 years. There were seven men and eight women. Clinical and treatment information was available for all patients (Table 1). All patients underwent either biopsy or tumor resection. Although treatments were variable, high-dose methotrexate (HD-MTX) was the most prevalent therapy used in combination with other treatment modalities.

Table 1 Clinical characteristics of patients with PCNSL
Full size table

Immunohistochemistry

Paraffin-embedded tissue sections were de-paraffinized and rehydrated using xylene and graded alcohols. Immunohistochemical staining was performed in a Dako Omnis autostainer (Dako North America, Inc. Carpinteria, CA, USA). The following primary antibodies were used: CD20 (L26), CD79a (12E7), CD10 (56C6), CD23 (DAK-CD23), BCL2 (124), BCL6 (PG-B6p), MUM1 (MUM1p), cyclin D1 (EP12), and Ki67 (MIB-1). The immunohistochemical staining was interpreted as positive or negative by a board-certified pathologist in all cases.

Targeted sequencing and tumor mutation burden

Formalin-fixed paraffin-embedded tumor samples were analyzed for genetic alterations by targeted NGS (FoundationOne™Heme, Foundation Medicine Inc., Cambridge, MA, USA). The FoundationOne® Heme assay was performed in a clinical laboratory improvement amendments (CLIA)-certified laboratory, as previously described [21]. Adaptor-ligated sequencing libraries were captured by solution hybridization with two custom bait-sets targeting 406 cancer-related genes, 31 genes frequently rearranged by DNA-seq, and 265 genes frequently rearranged by RNA-seq (Supplementary Table 1). The captured products were sequenced on HiSeq2500, Illumina. Sequenced data was evaluated for four classes of GAs: base substitution, copy number alterations, fusion/rearrangements, and insertion and deletions. Final NGS results were acquired from the patients’ FoundationOne Heme reports.

Tumor mutation burden (TMB) was calculated based on the number of somatic mutations in sequenced genes and extrapolating the value to the genome as a whole using a validated algorithm [22, 23]. TMB was reported as a number of mutations per megabase (mb) of genome. Based on the FoundationOne™Heme reports, TMB results were also categorized into three groups: low (1–5 mutations/mb), intermediate (6–19 mutations/mb), and high (≥ 20 mutations/mb). Values were rounded to the nearest integer.

Gene ontology

To do the gene ontology (GO) analysis, we downloaded GO gene sets from Molecular Signatures Database (MSigDB) [24]. A total of 5917 gene sets, including 4436 biological process, 580 cellular components, and 901 molecular function gene sets, were used in this study. The size of each gene set (N size) was calculated and the number of mutant genes in our dataset (50 genes) within each gene set was determined. To evaluate the enrichment of mutant genes in GO dataset, we computed the fraction of mutant genes in each gene set (50 genes/N size). All the analysis was done in R (version 3.4.2).

Results

Patient characteristics

Clinicopathologic characteristics for all patients are summarized in Tables 1 and 2. Four patients underwent resection; all others underwent needle biopsy. All tumors were positive for CD20 and CD79a by immunohistochemical staining. Employing the immunohistochemical algorithm of Hans et al. [25], all cases were sub-classified as either ABC (n = 7, 47%) or GCB subtype (n = 7, 47%), with the exception of one case for which immunohistochemical staining was not available. Both HIV positive patients were diagnosed with PCNSL of the ABC subtype. The results of cerebrospinal fluid (CSF) cytology were available for eight patients, all of them had a negative CSF-cytology result. Among nine CSFs tested by flow cytometry, four cases had a diagnosis of B-cell lymphoma, four cases were reported as non-diagnostic and one case was negative. One patient developed recurrence with diffuse osseous metastases. At the time of the study 2/15 (13.3%) patients had died of lymphoma, both patients had declined treatment and received comfort care only.

Table 2 Results of immunohistochemical stains for PCNSL
Full size table

Genomic alterations

A total of 79 GAs were detected in 50 genes (Fig. 1). The median number of GAs detected by the assay per patient was 5 (range 1–10). In 8 (10.1%) events, single genes harbored more than one GA. In the two HIV positive cases, only a single mutation was detected in MLL3 (KMT2C) and TUSC3 tumor suppressor genes. The rate of mutation frequency in HIV positive patients was lower than other PCNSLs. The most prevalent mutations were missense (n = 25, 32%) followed by frame shift (n = 18, 23%) (Fig. 1a). The most commonly mutated gene was MYD88 (n = 5, 33.3%), followed by alterations in CDKN2A/B (n = 5, 33.3%), and TP53 (n = 4, 26.7%). Fusion or rearrangement of the IGH gene with Bcl2, Bcl6, and MALT1 genes was also identified in 33.3% of cases (n = 5). It is worth to mention that all gene loss mutation was reported in CDKN2A/B gene (Fig. 1b). All cases with alterations in MYD88 (33.3%) had a missense mutation, the p.L265P substitution affecting exon 5, except one case that showed the less common p.V217F mutation. MYD88 mutations occurred more commonly in DLBCL of the ABC subtype (42.9%) versus the GCB subtype (28.6%). Of the 15 patients, 4 cases (26.7%) had high TMB (≥ 20 mutations/mb), 2 cases (13.3%) had an intermediate mutation burden (8–12 mutations/mb), and 9 cases (60%) had a low TMB (1–5 mutations/mb). Both HIV positive patients demonstrated a low TMB. The GO analysis revealed that genes mutated in PCNSL were enriched in pathways involved in “Cell Death”, “Positive Regulation of Transcription from RNA Polymerase II promoter”, “Mitotic DNA Integrity Checkpoint”, “Chromosome Organization”, “Cellular Response to Endogenous Stimulus”, and “Regulation of Smooth Muscle Cell Proliferation” (Fig. 1c).

Fig. 1
figure 1

Types of mutations observed in PCNSL. a Pie chart showing the percentages of different types of somatic mutations in PCNSL. Approximately, 50% of the mutations in PCNSL are missense or frameshift mutations. Gene rearrangements were detected in 10% of cases. b Genes mutated in PCNSL. Mutations in MYD88, CDKN2A/B, TP53 and IGH-BCL6 were the most frequent alterations detected. Asterisk indicates information from various sources (e.g., My Cancer Genome, COSMIC, Pubmed, and the Foundation Medicine report) were used to determine the predicted effect of the mutations on the function of each gene. c GO enrichment analysis of 50 mutated genes. The mutant genes were enriched in pathways of “Cell Death”, “Positive Regulation of Transcription from RNA Polymerase II promoter”, “Mitotic DNA Integrity Checkpoint” and “Chromosome Organization”

Full size image

Discussion

In the current study, we report the results of a comprehensive analysis of GAs in PCNSL, using a targeted NGS assay that evaluates more than 400 cancer-related genes. Among DLBCLs from different sites, PCNSL has the highest frequency of MYD88 mutations [26], which is in agreement with our results showing a high MYD88 mutation rate (33.3%) in PCNSLs. We also detected a higher frequency of MYD88 mutations in PCNSL of the ABC subtype, compared to the GCB subtype, in accordance with previous studies [12, 27]. Nonetheless, Fukumura et al. [3] reported no significant differences in MYD88 mutations between ABC- and GCB-lymphomas. Due to the low prevalence of PCNSLs, the majority of studies include a small cohort of patients, which can be partially responsible for variations in the reported frequencies of GAs. Although we report genomic mutations in a limited number of patients, our study includes evaluation for TMB and gene rearrangements, which have been neglected in the majority of previous studies on PCNSL. Previous studies have reported the absence of concurrent CXCR4 and MYD88 mutations in PCNSLs, in contrast to Waldenstrom macroglobulinemia, in which a significant number of patients show coexisting CXCR4 and MYD88 mutations [28, 29]. In our study, we identified one case of PCNSL with CXCR4 and MYD88 mutations. A recent study on genomic characterization of lymphomas also established the presence of the respective mutations in PCNSLs [30]. Waldenstrom macroglobulinemia patients who have MYD88 and CXCR4 mutations have been reported to be more resistant to ibrutinib treatment due to the activation of the AKT and ERK pathways [31,32,33]. The effects of CXCR4 mutations in PCNSL are currently unknown. However, an FDA-approved CXCR4 antagonist AMD3100 (Plerixafor/Mozobil) could potentially be a candidate therapy for patients with CXCR4 mutated malignancies [34, 35].

MYD88 mutations have been reported in other hematologic malignancies, including Waldenstrom macroglobulinemia, chronic lymphocytic leukemia (CLL) and mucosa-associated lymphoid tissue (MALT) lymphomas [12, 36]. In PCNSL, different studies have reported a wide range of mutation frequencies in the MYD88 gene ranging from 33 to 100% (Table 3). This wide spectrum is partially due to variations in sensitivity of the various assays utilized and differences in patient population. For instance, Hattori et al., reported the MYD88 p.L265P mutation was detected in 100% of cases (14/14) using ddPCR. In contrast, they could detect the mutation by targeted deep next generation sequencing (TDS) in 13 out of 14 cases [37]. In another study, whole exome sequencing revealed an MYD88 mutation in 75.6% of cases. However, after manual inspection of the negative cases, followed by sanger sequencing confirmation, the percentage or MYD88 mutant cases reached 85.4% [3]. Given the high prevalence of MYD88 mutations in PCNSLs, treatment with ibrutinib, which inhibits Bruton’s Tyrosine Kinase and further suppresses NF-κB and STAT3 activation and tumor growth, could be considered in patients with PCNSL [38, 39]. However, although there are ongoing clinical trials (NCT02315326), the efficacy of ibrutinib for the treatment of PCNSL remains to be determined. In addition to the therapeutic implication of the MYD88 mutation in PCNSL, its detection in cerebrospinal fluid has been recently introduced as a potential minimally-invasive approach for diagnosis [40].

Table 3 Reported mutations in PCNSL in the literature
Full size table

CDKN2A and CDKN2B encode p14ARF and p16INK4a, and p15INK4b tumor suppressor proteins, respectively. In accordance with our study, loss of CDKN2A/B genes have been commonly reported in DLBCL, which has been associated with reduced mRNA expression and higher tumor grades [45,46,47,48]. The p16INK4a and p15INK4b proteins maintain Rb tumor suppressor activity through suppression of CDK4 and CDK6 [49]. Therefore, using CDK4/6 inhibitors, including LY2835219, LEE011, and the FDA-approved palbociclib has been suggested as a potentially helpful therapy for CDKN2A/B mutated tumors [50]. However, initial clinical results did not provide promising results and further investigations for PCNSL is required [51,52,53,54]. The p14ARF protein is responsible for TP53 activation and induced cell death [55, 56]. TP53 was also among the commonly mutated genes in our study. A research study on 506 primary DLBCL patients treated with R-CHOP reported that TP53 mutation significantly correlate with worse survival in either ABC- or GCB-DLBCL [57].

There is a growing body of clinical and experimental evidence that TMB could be used as a biomarker for tumor prognosis and predicting response to immunotherapy [58, 59]. Various factors, including microsatellite instability [60], cigarette smoke [61, 62], and exposure to mutagens like UV light [63] have been associated with TMB in other tumors. High-TMB was reported in PCNSLs recently with a frequency of 33% [30], which is similar to our results (26.7%). From the therapeutic point of view, a high TMB was suggested to be associated with better prognosis in different tumors and susceptibility to nivolumab treatment in non-small cell lung cancer [58, 59]. A previous study analyzing 100,00 human cancer genomes showed that TMB determination, by comprehensive genomic profiling like the one reported in our study, correlated with TMB determined by whole exome sequencing [64]. This is clinically useful in regards to lymphoma, since it has been shown that high TMB and high PD-L1 expression in DLBCL may be linked to greater responsiveness to immunotherapeutic agents and checkpoint inhibitor therapies like anti-PDL1 [30, 65].

The presence of IGH-Bcl6 rearrangements in 20% of PCNSL is in accordance with prior studies (17–47%) [66,67,68,69]. Overexpression of Bcl6 due to the chromosomal translocation with impaired immunoglobulin IGH and further arrest in B-cell differentiation was reported to be one of the primary contributing factors to PCNSL pathogenesis. On the other hand, IGH-Bcl2 rearrangement, resulting from the reciprocal chromosomal t(14;18)(q31;q21) translocation was detected in one case (6.7%). Multiple studies have tried to define a prognostic role for Bcl2 and Bcl6 rearrangement in patients with PCNSL, however, the results are conflicting [67, 70,71,72]. A recent FDA approved anti-CD19 CAR T-cell therapy, Axicabtagene ciloleucel (KTE-C19) has been recommended for high-grade B-cell lymphoma with MYC, Bcl2 and/or Bcl6 rearrangement [73, 74]. However, its utility for the treatment of PCNSL remains to be determined. Characterization of GAs in PCNSL is critical for the development of non-invasive methods for diagnosis and targeted therapies [40]. In addition to point mutations, our study highlights other genomic events, including gene deletions, rearrangements or fusions in PCNSLs. While the number of patients in the study is limited, our findings broadens our understanding of the molecular heterogeneity of PCNSL.