Primary Central Nervous System (CNS) Lymphoma B Cell Receptors Recognize CNS Proteins Free

Primary lymphoma of the CNS (PCNSL) is a diffuse large B cell lymphoma (DLBCL) confined to the CNS (1, 2). Its exclusive manifestation in the CNS remains a mystery.

PCNSL cells correspond to late germinal center (GC) exit B cells with surface IgM expression (3, 4); they are characterized by rearranged and somatically mutated Ig genes with a high Ig gene-mutation frequency (5–8). Collectively, these features, which indicate participation of the tumor cells in a prolonged GC reaction, as well as the restricted Ig VH gene repertoire with a biased usage of the IGVH4-34 gene segment and evidence for ongoing somatic hypermutation (SHM), provide strong evidence for Ag selection for expression of a functional BCR (5, 8, 9). This has raised the intriguing hypothesis that Ags within the CNS may stimulate the tumor cells and may play a role in the pathogenesis of PCNSL similar to follicular lymphoma (FL), MALT lymphoma, and chronic lymphocytic leukemia (CLL), for which several Ags recently were identified (10–12). These recent observations may suggest antigenic stimulation as a common pathogenically relevant factor in the promotion of lymphomagenesis of mature B cell neoplasms.

To investigate the hypothesis that Ag recognition plays a role in the pathogenesis of PCNSL and to identify potential candidates, we cloned and expressed in vitro recombinant Abs (recAbs) exhibiting sequences identical to the variable part of the IgH and IgK/L chains of the tumor cell BCR in 23 PCNSLs. Although none of the PCNSL Igs was reactive with common Ags, the large-scale protein microarray, successfully used for autoimmune profiling in neurodegenerative and inflammatory disorders (13–15), identified 1547 proteins that were recognized by the 23 recAbs. Interestingly, recAbs frequently recognized proteins physiologically expressed by CNS neurons. In addition, recAbs, including all that were derived from IGHV4-34 using PCNSL, recognized galectin-3, which was upregulated on activated microglia/macrophages, astrocytes, and cerebral endothelial cells in the CNS invaded by PCNSL. Thus, CNS proteins may represent potential Ags that trigger the BCR of PCNSL cells, thereby fostering tumor cell survival and proliferation. These data emphasize the role of the microenvironment for PCNSL and may explain, at least in part, the exclusive manifestation of this DLBCL entity in the CNS.

Cryopreserved stereotactic PCNSL biopsies from 23 HIV⁻ patients were studied. Diagnoses were established according to World Health Organization classifications (1, 2). All samples had a tumor load ≥ 80%. Systemic lymphoma manifestation was excluded by extensive staging. The study was approved by local Ethics Commissions (06-187, 07-109) and performed in accordance with the Declaration of Helsinki.

High m.w. DNA was extracted from frozen tissue with a DNA extraction Kit (QIAGEN, Hilden, Germany), according to the manufacturer’s instructions.

Total RNA was extracted from frozen tissues with TRI reagent (Sigma-Aldrich, Taufkirchen, Germany). RNA was transcribed into cDNA using the QuantiTect Reverse Transcription Kit (QUIAGEN), according to the manufacturer’s instructions.

DNA was used to identify rearrangements of the IgH, IgK, or IgL loci using IGHV, IGKV, and IGLV family-specific primers, respectively, in a two-phase PCR approach with 20 cycles two times, as described (5). Primers hybridize to the leader peptide region of IGHV, IGLV, or IGKV.

Each PCR was analyzed by a QIAxcel Advanced Instrument (QIAGEN) using ScreenGel Software 1.2 and directly sequenced with the ABI PRISM Dye Terminator Cycle Sequencing Ready Reaction Kit v3.1 (Life Technologies, Darmstadt, Germany) on an ABI 3130 sequencer (Life Technologies), as described (5). Sequences were analyzed with 4Peaks Software v1.7.2 (The Netherlands Cancer Institute, Amsterdam, the Netherlands) and compared with human germline Ig gene sequences in the International ImMunoGeneTics database (16).

cDNA was used for amplification and cloning of the sequences of the V region genes, as described (17). Cloning vectors were kindly provided by Dr. Hedda Wardemann (Max Planck Institute for Infection Biology, Berlin, Germany) and Dr. Michel C. Nussenzweig (Rockefeller University, New York, NY).

For expression of recAbs, serum-free 293FT cells were transfected with two expression vectors (IgH and IgK or IgL); 5 d later, supernatants were harvested. After Ab-specific purification of supernatants using protein A columns (Merck Millipore, Schwalbach, Germany), eluates were concentrated with a 100-kDa filter membrane using Amicon Ultra-15 100-kDa filter tubes (Merck Millipore) to a final volume of 200 μl. Purity of recAb was determined on a denaturing gel stained with Coomassie.

In addition to recAbs derived from PCNSL, five recAbs derived from FL (513, 526, 684, 1081, 1102) were included (18). They were kindly provided by Dr. Ronald Levy and Dr. Debra K. Czerwinski (Department of Medicine, Division of Oncology, Stanford University School of Medicine, Stanford, CA).

Correct folding and potential functionality of recAbs were determined by a sandwich ELISA with monoclonal mouse anti-human IgG (clone GG-6; Sigma-Aldrich) as catch Ab and alkaline phosphatase–coupled mouse anti-human IgG (clone GG-5; Sigma-Aldrich) as detection Ab; human IgG (Sigma-Aldrich) served as standard. ELISA was detected with p-nitrophenyl-phosphate (Sigma-Aldrich).

recAbs were tested for potential autoreactivity by ANAcombi and ANCAcombi ELISA (Orgentec, Mainz, Germany).

recAbs were tested for potential autoreactivity by indirect immunofluorescence assays on HEp-2 cells using the ANA-HEp-2 Kit (Orgentec), according to the manufacturer’s instructions.

Protein microarray slides (ProtoArray; Life Technologies) stored at −20°C were equilibrated at 4°C and at room temperature for 15 min each. Blocking, Ab incubation (dilution in washing buffer 1:5000), and washing steps were performed according to the manufacturer’s protocol. After processing, protein microarrays were scanned (Array-Scanner FR202; Strix Diagnostics, Berlin, Germany); the images were analyzed to acquire fluorescence intensity raw data as gpr files (StrixAluco 3.0 software; Strix Diagnostics).

After raw data quality management and preprocessing, M Statistic (“M Score”)–based group comparisons were performed using Prospector software (v5.2; Life Technologies) to select candidate proteins.

To control for specificity of Ab binding, cdk2 (mouse IgG1, clone D-12; Santa Cruz Biotechnology, Heidelberg, Germany) and galectin-3 (rat IgG2a, clone M3/38; BioLegend, Fell, Germany) were used as Abs with defined specificity. They were detected with goat anti-mouse IgG Alexa Fluor 647 conjugate and goat anti-rat IgG Alexa Fluor 647 conjugate (both from Life Technologies). As expected, they identified their targets as evidenced by high signal intensity (data not shown).

A total of 0.1 or 0.5 μg recombinant GRINL1A (OriGene Technologies, Rockville, MD) was separated on 4–12% Bis-Tris protein gels and transferred to nitrocellulose membranes, according to the manufacturer’s protocol (Life Technologies). Blots were blocked using StartingBlock Blocking Buffer (Thermo Scientific, Schwerte, Germany) prior to probing with the recAb and anti-human IgG IRDye 800 (Rockland, Gilbertsville, PA) as a secondary detection Ab. Scanning was performed using an Odyssey Classic scanner (LI-COR Biosciences, Bad Homburg, Germany). After stripping the membranes using Restore Plus Western blot Stripping Buffer (Thermo Scientific), equal loading of GRINL1A was controlled by probing with anti-DDK Ab (OriGene Technologies) and IRDye 800CW donkey anti-mouse IgG (LI-COR Biosciences).

The expression of galectin-3 protein (LGALS3) was studied in an independent series of 20 paraffin-embedded PCNSLs from HIV⁻ patients (15 female, 5 male; mean age: 70 y; median age: 73 y; range: 45–86 y). Stereotactic brain biopsies without alterations, with minor unspecific inflammatory and reactive changes, and with inflammatory demyelination were included as controls. Immunohistochemistry was performed on 4-μm paraffin sections with monoclonal mouse anti-galectin-3 (NCL-GAL3, clone 9C4; Novocastra; Leica, Wetzlar, Germany), with 3,3-diaminobenzidine as chromogen and H₂O₂ as cosubstrate on an automated immunostainer (Bond; Leica). Sections were counterstained slightly with hemalum.

Immunoreactivity of recAbs 03 and 23 was studied by double immunofluorescence using the respective recAb with mouse anti-human galectin-3 Ab (NCL-GAL3, clone 9C4; Novocastra; Leica) in galectin-3–expressing PCNSL, inflammatory CNS disorders, and normal brain derived from brain biopsies adjacent to low-grade glioma. Studies were performed on 4-μm paraffin sections. After pretreatment with Ag-retrieval solution at pH 9 (DCS, Hamburg, Germany) for 20 min, sections were incubated with the respective recAb labeled with green fluorescent dye CF488A using the Mix-n-Stain 488A Ab Labeling Kit (Sigma-Aldrich, St. Louis, MO), according to the manufacturer’s instructions. Thereafter, monoclonal mouse anti-galectin-3 was applied, followed by a species-specific Ig coupled with DyLight 549 (red) (Dianova, Hamburg, Germany).

Gene expression profiles (GEPs) of 10 normal CNS tissues and 21 PCNSLs published recently (3) were analyzed for the expression of mRNA corresponding to proteins identified in the ProtoArray studies.

Biopsies from 23 HIV⁻ patients (12 females, 11 males; mean age: 69 y; median age: 71 y; range: 28–86 y) were studied. All tumors were classified as DLBCL of the CNS according to World Health Organization classification (1, 2). The tumor cell phenotype was CD20⁺BCL6⁺MUM1⁺ with a high mitotic and proliferative activity (>70% or even >90% MIB-1⁺ cells).

PCR analysis detected a monoclonally rearranged IGH gene together with either a monoclonally rearranged IGK or IGL gene in each PCNSL. Corresponding IGHV and IGKV or IGLV gene segments were identified in all cases (Table I). In accordance with previous studies (5–8), genes of the IGHV4 family were rearranged most frequently (12/23, 52%), with a preference for the IGHV4-34 gene segment (6/12, 50%); thus, IGHV4-34⁺ PCNSLs accounted for 26% of tumors of this series. In addition, 8 of 23 (35%) PCNSLs had rearranged a gene segment of the IGHV3 family. Two and one PCNSLs used IGHV5-51 and IGHV1-46 (Table I), respectively. All PCNSLs had introduced somatic mutations into their rearranged IGHV, IGKV, or IGLV genes. The mean mutation frequency was 6.9 and 7.9% for IGHV and IGK/LV genes, respectively. Mutations corresponded to base pair exchanges, whereas deletions, insertions, and stop codons were absent. Each rearrangement was in the correct reading frame and did not harbor a stop codon; thus, all PCNSLs harbored IGHV and IGK/LV rearrangements encoding a potentially functional BCR.

To address BCR reactivity of PCNSL cells, we cloned and expressed in vitro recAbs from 23 PCNSLs. For all recAbs generated by RT-PCR, correctness of sequences was confirmed by sequencing and comparison of the sequences amplified from cDNA and DNA isolated from the same PCNSL specimen. Furthermore, their potentially functional folding was demonstrated (Fig. 1). Thus, the RT-PCR–based technique to produce recAbs with BCR specificity of PCNSL cells was shown to be feasible and efficient.

Testing for potential reactivity with common autoantigens revealed that none of the 23 recAbs exhibited autoimmune reactivity with RNP-70, RNP/Sm, Sm, SS-A, SS-B, Scl-70, centromere B, Jo-1, PR3, MPO, BPI, elastase, cathepsin G, lysozyme, or lactoferrin. Furthermore, recAbs did not show nuclear, subnuclear plus cytoplasmic, or cytoplasmic reactivity with HEp-2 cells (data not shown), which are frequently used to identify Ags recognized by autoantibodies in patients with autoimmune disorders.

Thus, these data suggest that tumor Igs from PCNSL are not reactive against self-Ags commonly involved in autoimmune disorders.

All recAbs specifically recognized proteins present on the ProtoArray. Overall, 1574 proteins were recognized by at least one Ab (http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE61556) (19). When considering all 23 recAbs, they recognized a mean of 200 proteins (median: 96; range: 62–818). For comparison, five recAbs derived from FL were included in the ProtoArray analysis. Overall, they recognized 606 proteins in total, with a mean of 330 proteins/recAb (median: 330; range: 257–403). Together, these data indicate that mature B cell lymphomas of the PCNSL and FL type react with a large number of proteins.

recAbs derived from IGHV4 using PCNSL (n = 12) reacted with a mean of 244 proteins (median: 124; range: 66–818). With regard to IGHV3-expressing PCNSL (n = 8), a mean of 181 proteins was recognized (median: 95; range: 62–580). Two IGHV5 PCNSL-derived recAbs recognized 90 proteins (range: 69–70), whereas the Ab derived from the only IGHV1-expressing PCNSL recognized 73 proteins (Fig. 2). Considering recAbs from all IGHV families used by the PCNSL of this cohort, 55 proteins were recognized by PCNSLs of all IGHV gene families. recAbs derived from the dominant IGHV gene families in PCNSL (i.e., IGHV3 and IGHV4) recognized 553 proteins (Fig. 2). Thus, these data indicate polyreactivity of tumor Igs.

Considering IGHV4-34–using PCNSLs, the respective recAbs recognized a mean of 391 proteins (median: 236, range: 96–818). Comparison of recAbs derived from six IGHV4-34–using PCNSLs with the 17 other PCNSLs (not using IGHV4-34) revealed that 651 of the total 1574 proteins recognized by the recAbs were bound by Abs derived from IGHV4-34–using PCNSLs only. The majority of proteins recognized (1097) is attributed to recAbs derived from IGHV4-34–using PCNSLs, because non–IGHV4-34–derived recAbs recognized only 67 proteins.

In additional analyses, annotated proteins, which are secreted, membrane bound, or expressed on the cell surface and recognized by ≥50% of the recAbs in all PCNSLs and in the predominant subgroups (i.e., IGHV3, IGHV4, and IGHV4-34), were selected from the 1547 proteins recognized on the ProtoArray. We also eliminated Igs to exclude potentially homotypic aggregation of Abs distinct from specific BCR binding. This procedure narrowed down the large number of proteins recognized by the recAbs to proteins capable of interaction with the BCR of PCNSL tumor cells and that are recognized by a significant number of recAbs. Selecting for annotated and secreted proteins yielded 21 proteins (Table II). LGALS3, LGALS8, ICAM-4, HSH2D, KCNAB1, P2RY2, and PDDC1 were the proteins recognized by recAbs of all IGHV gene families (Table II). Furthermore, recAbs 526, 684, and 108, but not recAb 513, from FL, also recognized LGALS3 (data not shown).

Western blot analysis confirmed the ProtoArray data as evidenced as proof-of-principle for GRINL1A. All five recAbs that recognized GRINL1A on the ProtoArray bound to GRINL1A in a concentration-dependent manner on Western blot (Fig. 3).

For 13 of the 21 proteins of interest (Table II), corresponding 18 mRNA tags were present on the Affymetrix U95Av2 chip. Generally, expression of these mRNAs was virtually complementary in PCNSL and normal CNS tissue (Fig. 4). Interestingly, LGALS3 mRNA was expressed at increased levels compared with normal brain samples (Fig. 4). These mRNA data are extended by studies at the protein level. Immunohistochemistry demonstrated that galectin-3, which is not expressed in normal CNS tissue, is induced under pathological conditions, including inflammation and tumor infiltration (20, 21). In the brain invaded by PCNSL, tumor-infiltrating macrophages and activated resident brain cells (i.e., astrocytes, microglia, and cerebral endothelial cells) expressed galectin-3 (Fig. 5). In this regard, activated brain cell populations responded similarly to inflammatory conditions (Fig. 5). Interestingly, cells from 5 of 20 (25%) PCNSLs also expressed galectin-3 (Fig. 5).

Double immunofluorescence studies further confirmed the reactivity of recAbs 03 and 23 with galectin-3 by costaining with a commercially available galectin-3 Ab (clone 9C4), labeling the cytoplasm of tumor cells in galectin-3–expressing PCNSL (Fig. 6A, 6B). In addition, both recAbs reacted with the nuclei of the tumor cells, which indicates their polyreactivity. In inflammatory CNS disorders, recAbs 03 and 23 bound to macrophages, activated microglial cells, and reactive astrocytes, colocalizing with mouse anti-human galectin-3, clone 9C4 (Fig. 6C, 6D). In addition, both recAbs also reacted with small, mature lymphocytes, further indicating polyreactivity. In normal brain adjacent to various pathologies, recAb 03, recAb 23, and mouse anti-human galectin-3, clone 9C4 reacted with cerebral endothelial cells (Fig. 6E, 6F). In addition, recAb 23 reacted with oligodendrocytes, further indicating polyreactivity (Fig. 6F).

This study demonstrates that the BCR of PCNSL frequently recognizes proteins that are physiologically expressed in the CNS. This observation supports the hypothesis that antigenic stimulation may support tumor cell proliferation and survival and, moreover, may contribute to the confinement of PCNSL to the CNS, suggesting a possible role for CNS Ags as microenvironmental factors contributing to the pathogenesis of PCNSL.

In accordance with previous studies, Ig gene analysis demonstrated the introduction of somatic mutations into the rearranged IGHV and IGK/LV genes [this study and (5–8)]. The mutational pattern shows counterselection of replacement mutations in the framework regions. Furthermore, PCNSLs show intraclonal diversity resulting from ongoing SHM (5, 8). Collectively, there is strong evidence for selection of PCNSL cells by Ag(s) (5, 6, 8, 9), suggesting that Ag recognition plays a role in the pathogenesis of PCNSL. Furthermore, the intriguing question arises as to the existence of extranodal SHM in the CNS, which is of fundamental immunological relevance beyond PCNSL.

To identify potential Ags that might trigger the BCR, we cloned and expressed recAbs that are characterized by their binding specificity identical to the BCR of the tumor cells in a cohort of 23 PCNSLs. Taking advantage of these recAbs, we studied their binding profile, including a search on a large-scale protein microarray containing 9500 protein spots. Remarkably, 1547 proteins were recognized by at least one of the 23 recAbs of this cohort of PCNSLs, with a mean of 200 proteins recognized. For comparison, five recAbs derived from FL recognized a high number of proteins (606 in total), with a mean of 330 proteins/recAb, indicating polyreactivity of both lymphoma types. Zhu et al. (22) studied BCR reactivity in ocular adnexal extranodal marginal zone lymphoma, also taking advantage of the ProtoArray. Their use of additional selection strategies revealed a mean of 16 proteins identified by a recAb. Although this selection procedure is not a conditio sine qua non for analysis of ProtoArray raw data, application of these criteria (22) to PCNSL and FL reduces the numbers of proteins markedly. Nevertheless, the identification of a mean of 35 (median: 36) proteins in PCNSL and a mean of 87 (median: 87) proteins in FL recognized by the recAb still demonstrates that the ProtoArray data are indicative of polyreactivity. Irrespective of the stringency criteria applied for identification of hits on the ProtoArray as proteins bound by the recAb, these data indicate that polyreactivity may be a feature common to mature B cell lymphomas, including PCNSL, FL, and ocular adnexal extranodal marginal zone lymphoma (22).

Among the large group of 1547 proteins recognized, 55 proteins were recognized by at least one recAb of all IGVH gene families, thus demonstrating an overlap in recognition, irrespective of the IGHV gene family used. In addition to proteins recognized by Igs of all IGHV gene families, proteins were recognized by individual recAbs derived from different IGHV gene families. These data indicate the existence of Ags shared among tumors, as well as individual Ags specifically recognized by individual PCNSLs.

recAbs derived from IGHV4-34–using PCNSLs recognized the largest number of proteins, even when the 6 IGHV4-34–using PCNSL were compared with the remaining 17 PCNSLs not using IGHV4-34. This observation may further support the notion that the presence of a rearranged IGHV4-34 gene segment, which shows a biased usage and has been noted in 26–80% of PCNSLs [this study and (5–8)], may identify a distinct subgroup of PCNSL. A study specifically contrasting the molecular and clinical features of IGHV4-34 with IGHV-non4-34 PCNSLs has not been performed. Interestingly, IGHV4-34–encoded Abs are elevated in the serum of patients with systemic lupus erythematosus and infectious mononucleosis (23), which prompted us to search for autoreactivity of PCNSL Igs. However, none of our recAbs reacted with common autoantigens. In this regard, PCNSL apparently differs from other types of lymphoma. Reactivity with HEp-2 cells [in a frequency similar to normal memory B cells (24, 25)] identified 26% of FLs, which recognized self-Ags (18). Moreover, IgG FL exhibited a higher frequency of self-reactivity than did IgM FL (20.4% versus 3.4%) (10). One may suggest that this fact also accounts, at least in part, for the difference from PCNSL, which retains IgM expression due to impaired class switch recombination (4). Furthermore, vimentin was identified as a shared autoantigen recognized by 19, 50, and 33% of Igs of FL, mantle cell lymphoma, and multiple myeloma, respectively (10). In our cohort of PCNSLs, only two recAbs recognized vimentin. Collectively, these observations may suggest the existence of shared autoantigens across mature B cell lymphomas.

In CLL, highly polyreactive Abs are expressed by unmutated CLL B cells. In addition, reversion of mutated CLL Ab sequences to their germline counterparts encoded polyreactive and autoreactive Abs (11), suggesting that self-reactivity may be altered by SHM and that shared Ags may select tumor cell progenitors. Although SHM may alter tumor Ig reactivity, PCNSL differs from CLL by showing polyreactivity despite being mutated. In addition to self-Ags, MALT lymphoma was reported to bind bacterial Ags (26). For PCNSL, a role for infectious agents has not been delineated. Efforts focusing on viruses with the propensity to persist in the CNS excluded EBV, HHV6, HHV8, and SV40 as potential candidates (27–29).

Focusing on proteins of the ProtoArray that qualify for specific BCR interactions by selecting for annotated proteins that are secreted, membrane bound, or expressed on the cell surface with repeated recognition by the PCNSL Igs identified 21 proteins. Despite the fact that individual PCNSLs may recognize unique proteins that may have escaped detection under the selection conditions defined in this study, our selection strategy aimed to identify proteins that may be involved in the interaction with the BCR of PCNSL cells. The identification of 21 proteins instead of a single protein adds to the notion that PCNSLs do not show a restricted CDR3 length and lack stereotyped BCR (6); together, these observations favor involvement of more than one Ag stimulating the tumor cell BCR.

GRINL1A, ADAP2 (centaurin-α), and BAIAP2, which are among these 21 proteins, are physiologically expressed by CNS neurons and involved in neuronal excitability, neurotransmission, and cytoskeletal cross-talk. These proteins are altered in Alzheimer’s disease and attention deficit/hyperactivity disorders (30). Thus, neurons may provide antigenic signals for the BCR of PCNSL via GRINL1A, centaurin-α, and BAIAP2. Under pathological conditions, including tumor infiltration and inflammation, endoglin (ENG in Table II) and galectin-3 are differentially induced on resident brain cells and may provide further antigenic targets for the BCR of PCNSL cells on cerebral endothelial cells (endoglin, galectin-3), as well as on activated astrocytes, microglia/macrophages, and ependymal cells (galectin-3) (20, 31).

Thus, resident CNS cell populations, including neurons, astrocytes, ependymal cells, microglia/macrophages, and cerebral endothelial cells, are able to interact with the BCR of malignant B cells of PCNSL in a cell type–specific manner. These interactions may also contribute to the characteristic topography of PCNSL, which preferentially affects the basal ganglia and periventricular tissue. From a functional point of view, this interaction of CNS proteins with the BCR of the tumor cells may foster BCR signaling in PCNSL, similar to mutations in proximal BCR components like CD79B and CARD11, which are present in 20 and 19% of PCNSLs, respectively (32, 33), ultimately sustaining survival and proliferation via continuous NF-κB activity (34). These data further stress the importance of BCR transmission of a tonic survival signal that usually is enhanced upon BCR binding to its cognate Ag and support the hypothesis of Ag-driven growth of PCNSL cells.

The role of galectin-3 in PCNSL may reach beyond BCR triggering. In 5 of 20 (25%) PCNSLs analyzed in this study, the tumor cells expressed galectin-3, as shown by immunoreactivity with mouse anti-human galectin-3, clone 9C4, as well as by costaining with recAbs 03 and 23. Galectin-3 expression by PCNSL cells is in accordance with a study by D’Haene et al. (35), who identified its expression by the tumor cells of 8 of 46 (17%) PCNSLs, including one immunocompromised patient. In this subset, tumor cell galectin-3 may bind to CNS glycans, including Mac1, Mac3, and CD45, which are expressed on microglia/macrophages, and to CD45, which is expressed on reactive T and B cells in the inflammatory infiltrate associated with PCNSL. In PCNSL, galectin-3 may also promote resistance to apoptosis via homotypic binding to CD45, similar to systemic DLBCLs, 65% of which express galectin-3 (36–40).

Microenvironmental factors serving as self-Ags also were identified for FL, in which 19% of tumor Igs recognized vimentin expressed in the lymph node and one tumor in which myoferlin served as self-Ag (10, 18). However, for the immunologically privileged CNS, microenvironmental proteins serving as potential Ags for B cells have not been described before. The identification of potential Ags that may stimulate the BCR of tumor cells bridges the gap in the basal requirements for extranodal SHM in the CNS. Although APCs (microglia/macrophages) and T cells are well-known constituents of the reactive inflammatory infiltrate, BCR-stimulating Ag(s) have not been identified previously. In contrast, it also is conceivable and not mutually exclusive that the tumor cells may have experienced a GC reaction in secondary lymphatic organs, with subsequent homing to the CNS where they have encountered the same Ag or an Ag with shared epitopes that also triggers their BCR. The latter scenario is similar to autoimmune inflammation of the CNS, for which molecular mimicry was shown to underlie CD8 T cell–mediated autoimmune reactions directed against neurons and oligodendrocytes, which require the presence of the (auto)antigen recognized by the Ag-specific CD8 T cell in the brain (41, 42).

For polyreactivity, SHM may play a crucial role. The tumor cells of PCNSL are selected for a functional BCR by SHM. Furthermore, they are unable to terminate the GC program, as shown by ongoing SHM (5, 8). Moreover, aberrant SHM is active in PCNSL, and there is evidence that it plays a major role in the pathogenesis of PCNSL (5, 43). Nevertheless, there is a strong selection for a functional BCR (6) and, thus, for fostering signals through the BCR pathway (32). Altogether, this scenario indicates that PCNSL cells are polyreactive as a result of continued signaling through the BCR pathway because of their inability to terminate the GC program.

Thus, PCNSL cells, which are still selected for expression of a functional BCR, may interact with proteins expressed in the CNS via their highly mutated and polyreactive BCRs, ultimately creating a fertile microenvironment and sustaining tumor cell survival in the CNS. These data further support targeting the BCR pathway in PCNSL as a promising therapeutic strategy.

We thank Dr. Ronald Levy and Dr. Debra K. Czerwinski for kindly providing recAb from FL as controls. The technical assistance of Elena Fischer, Mariana Carstov, Diana Rudakova, Kathrin Barlog, and Jale Stoutjesdijk is gratefully acknowledged.

The authors have no financial conflicts of interest.