Abstract
Purpose: Only a limited number of genetic studies have been conducted in primary central nervous system lymphomas (PCNSL), partly due to the rarity of the tumors and the very limited amount of available tissue. In this report, we present the first molecular characterization of copy number abnormalities (CNA) of newly diagnosed PCNSL by array-based comparative genomic hybridization (aCGH) in formalin-fixed paraffin-embedded (FFPE) specimens and compare the results with matched, frozen tumor specimens.
Experimental Design: We conducted aCGH in FFPE tissues from PCNSL. Results were compared with matched, paired, frozen tumors.
Results: Our analysis confirmed the good to fair quality and reliability of the data generated from limited amounts of tumoral FFPE tissue. Overall, all PCNSL cases were characterized by highly complex karyotypes, with a median of 23 CNAs per patient (range, 17–47). Overall, 20 chromosomal regions were recurrently found in more than 40% of cases. Deletions of 6p21, 6q, and 9p21.3 and gain of 12q12-q24.33 were the commonest CNAs. Other minimal affected regions were defined, and novel recurrent CNAs affecting single genes were identified in 3q26.32 (TBL1XR1) and 8q12.1 (TOX).
Conclusions: The results obtained are encouraging. Larger archival tissue collections can now be analyzed to complement the still fragmented knowledge we have of the genetic basis of the PCNSL. Clin Cancer Res; 17(13); 4245–53. ©2011 AACR.
Translational Relevance
Genomic analysis has been challenging in primary central nervous system lymphomas (PCNSL), mainly due to the rarity of the tumors and the very limited amount of available tissue. In this study, we used tissue extracted from formalin-fixed paraffin-embedded (FFPE) blocks to conduct array-based comparative genomic hybridization (aCGH) assays in PCNSL. The good quality of the results is encouraging. Larger archival tissue collections can now be analyzed to complement the still fragmented knowledge we have of the genetic basis of PCNSL, which is based on a very few studies conducted in small cohorts. In addition, we used an aCGH platform that provided us with the highest resolution analysis of the PCNSL genome conducted to date and it enabled us to identify novel recurrent abnormalities affecting potential key genes in PCNSL pathogenesis.
Introduction
Primary central nervous system lymphoma (PCNSL) is an aggressive primary brain tumor characterized by a perivascular accumulation of malignant cells that have lymphoid characteristics. Its cell of origin and tumorigenesis are unknown. PCNSL is typically lethal without treatment, is increasing in incidence, and targets vulnerable populations (1, 2). It is therapy responsive, and its aggressive management may lead to remission (3, 4). However, current treatments are not increasing cure rates (5) and the quality of such survival is often poor (6, 7). PCNSL has a disproportionate effect on quality of life because of its disabling impact on cognition, language, mobility, and adaptive skills. The advanced age of the average patient and neurotoxicity of standard therapy further amplify this morbidity.
The 2000 NCI-NINDS PRG report stated that “Molecular characterization of PCNSL tumorigenesis is now needed to inform pathogenesis-based treatment and prevention strategies” (8). To that end, a series of gene function–oriented publications including our own have identified chromosomal abnormalities that may have pathogenetic relevance (9). For example, deletions of 6q are a frequent observation in systemic diffuse large B-cell lymphoma (DLBCL) and are typically associated with a worse prognosis (10, 11). Cady and colleagues described a similar association of 6q22–23 deletions and prognosis in PCNSL patients with no apparent immunodeficiency and stated that the deletion implied a loss or modification of a tumor suppressor gene, PTPRK (11). In contrast, posttransplant lymphoproliferative disorder of the central nervous system (CNS), the Epstein–Barr virus (EBV)-driven PCNSL-like disease seen in organ transplant recipients and other immunosuppressed states, is not associated with deletions of 6q or abnormalities of c-MYC and BCL-6, suggesting a distinct pathogenesis (12).
Unlike systemic DLBCL and other forms of non–Hodgkin's lymphomas, relatively little is known about the biology of PCNSL. Only a limited number of genetic studies have been conducted, partly due to the rarity of the tumors (3% of all primary brain tumors; ref. 13) and the very limited amount of available tissue. PCNSL lesions are typically deep-seated and usually best approached by stereotactic techniques (14). Aggressive debulking does not improve prognosis (13, 14), thus limiting the amount of tissue that can be safely removed, and specimens are typically exhausted by the diagnostic workup. In addition, no cell lines have been established in PCNSL, thus making the selection of appropriate in vitro systems for genomic validations and functional analysis more difficult. In PCNSL, the source of biological samples is often limited to formalin-fixed and paraffin-embedded (FFPE) specimens. These specimens represent valuable materials for cancer research, especially in retrospective studies with long follow-up data, but its use in systematic studies has been challenging because the yield of DNA obtained from this source is often very degraded. However, in recent years, there has been progress in the use of FFPE specimens for whole-genome array-based studies in solid tumors (15–17).
In this report, we describe the molecular characterization of copy number abnormalities (CNA) by array-based comparative genomic hybridization (aCGH) in matched FFPE and frozen tumor specimens taken from newly diagnosed PCNSL patients without apparent immunodeficiency. The aim of this study was, first, to better characterize the disease at chromosomal and gene levels and to correlate these features with clinical and pathologic features and, second, to validate the use of aCGH in FFPE tumor specimens as an alternative to frozen specimens.
Patients and Methods
Tumor samples
Frozen PCNSL tumor samples in pellets and FFPE blocks were retrieved from the Mayo Clinic Tumor Registry (Mayo Foundation IRB approval 08-001933) and the University of Virginia (University of Virginia IRB approval 14225). A review of clinical histories confirmed that each case was newly diagnosed, was confined to the CNS, and had no occult disease by standard staging and that each patient had no apparent immunodeficiency. Given the limited number of patients and the pilot nature of the study, statistical analyses were not done.
Three 3-μm sections were cut from each pellet and placed on clean glass slides. One slide was stained with hematoxylin–eosin (H&E) to confirm sufficient tumor-rich tissue without significant hemorrhage, necrosis, or artifact. Confirmatory immunohistochemistry (IHC) using antibodies directed against CD20 and CD3 was done on the second and third slides of each patient set. All confirmed cases were then screened for EBV, utilizing in situ hybridization probes that recognize EBV-encoded RNA (EBER). Any nuclear staining in tumor cells was viewed as a positive result. The specimens were subgrouped as germinal center (GC) or non-GC by IHC for CD10, MUM-1, and BCL-6, according to the Hans algorithm (18). Immunostaining was done in 1 batch to maximize laboratory efficiency. Each batch contained positive and negative controls and replicates from each case. Lymphomas were considered positive if 30% or more of the cells stained with antibody. Data on the intensity of staining were not used because of potential variations in tissue fixation and processing.
DNA isolation
Genomic DNA was obtained from frozen tumors by using the Puregene Core A Kit (Qiagen) according to manufacturer's recommendations. In the FFPE specimens, a slide was cut and stained as aforementioned to confirm the richness of tumors. In cases for which tumor burden was less than 70%, tumor areas were identified on the H&E-stained section. These areas were macrodissected to minimize normal tissue content. Samples were deparaffinized using heptane at room temperature for 1 hour. Methanol was added and the sample was pelleted by centrifugation. DNA was obtained using the AllPrep DNA/RNA FFPE Kit (Qiagen) according to manufacturer's recommendations. DNA concentration and purity were measured by spectrophotometry, and DNA integrity was assessed on a 1% agarose gel.
Array-based comparative genomic hybridization
To assess the reliability of FFPE samples, we conducted a comparison of assay performance between snap-frozen and FFPE samples in 6 PCNSL specimens. In an additional case, only the frozen tumor sample was analyzed.
Snap-frozen samples
aCGH was done in 3 samples with the Human Genome 244A microarray and in the remaining 4 with the Human Sureprint G3 microarray (Agilent Technologies). The digestion, labeling, and hybridization steps were done as previously described (19). Microarrays were scanned in a DNA Microarray scanner (Agilent Technologies). Feature extraction was done with Feature Extraction Software (version 9.5; Agilent Technologies). Extracted data were imported and analyzed using Genomic Workbench (version 5.0.14; Agilent Technologies) and Nexus 5.1 (Biodiscovery).
FFPE samples
aCGH was done on 6 FFPE specimens, using the Human Genome 244A microarray. Briefly, 2 μg of reference DNA was fragmented by heating at 95°C for 10 minutes. Fragmentation was not necessary for the FFPE samples. Samples were labeled with the ULS kit (Agilent) for FFPE tissues according to manufacturer's recommendation. Labeled genomic reactions were cleaned up with KREApure columns (Agilent Technologies) and hybridized at 65°C for 40 hours. Scanning and data analysis were the same as those for the snap-frozen samples.
Data analysis
Copy number abnormalities (CNAs) were calculated using aberration detection module 1 and RANK segmentation algorithms in Genomic Workbench and Nexus software, respectively (20). The derivative of the log ratio spread (DLRSpread) across the entire genome was calculated and used as a surrogate of assay quality. Assays with DLRSpread values lower than 0.2 are considered excellent; values between 0.2 and 0.3 are considered good; and values higher than 0.3 are considered marginal. Both a 2-probe filter (0.25_log2; 244K array format) and a 3-probe filter (0.25_log2; Sureprint G3 format) were used in the aberration detection of fresh samples, obtaining an average genomic resolution of 17 and 4.5 kb, respectively. Copy number variations were identified and excluded from the analysis as previously described (19).
FISH
Interphase FISH for 6q22 and 9p21 loci was done using custom DNA probes as previously reported (11) or, if one existed, a commercially available probe was used according to the manufacturer's instructions. A minimum of 50 tumor cells were scored. A cohesive group of at least 20 cells, of which at least 80% were abnormal, was required for the sample to be considered abnormal.
Results
Clinical characteristics
The cohort comprised 7 patients, 2 men and 5 women, with median age of 60 years (range, 52–77 years). All cases were histologically reconfirmed as CD20-positive DLBCL. Each specimen had diagnostic tissue without hemorrhage or necrosis; and each specimen was EBER negative. Three patients were identified at Mayo Clinic and 4 at the University of Virginia. No patient was HIV positive, had received a solid organ transplant, or had an active autoimmune disorder. Clinical details are shown in Table 1. Performance score could be ascertained in 5 patients. Survival data were available for all 7 patients (cases A–G) and treatment information for 5 patients (cases A–D and G). Treatment varied from patient to patient. Median survival of the 5 treated patients was 12 months. Two patients received no treatment and survived 2.5 months and 3 weeks, respectively (cases B and D).
Clinical characteristics of the patients included in this study
Of the 7 PCNSL patients, 3 were classified as having a GC phenotype and the remaining 4 as non-GC phenotype (ABC). In 6 of these 7 PCNSL patients (cases B–G), aCGH experiments were conducted on DNA samples obtained from both snap-frozen and FFPE tissues for assay quality comparative purposes. In the remaining sample (case A), only frozen sample was analyzed.
Performance of aCGH experiments in samples from FFPE and snap-frozen specimens
To measure the assay quality of the aCGH experiments, we used the DLRSpread value as a surrogate of signal noise. All aCGH experiments run on snap-frozen tumor specimens showed DLRSpread values in the range of good or excellent, based on QC thresholds obtained from Genomic Workbench software. On the other hand, experiments run on FFPE tumor specimens showed higher DLRSpread than paired frozen samples, with 3 assays considered in the good range and another 3 in the marginal range (Fig. 1A).
A, assay quality comparison between matched paired FFPE and snap-frozen samples. Case A was analyzed only from snap-frozen sample. B, correlation between assay quality (DLRSpread used as a surrogate) and the CNA concordance level between matched paired FFPE and snap-frozen samples. DLRSpread values below 0.2 correspond to excellent quality assays; values between 0.2 and 0.3 correspond to good quality; and values above 0.3 are considered marginal. C, the concordance of CNA detection between paired samples increases with the size of the abnormalities; D, graph showing that higher-quality assay is associated with higher concordance of CNA breakpoint location between paired samples.
Next, we analyzed the reliability of the CNA detection in assays run on FFPE specimens compared with the paired frozen samples. The data showed an inverse correlation between DLRSpread values and the concordance of aberration calls between pairs of samples (Fig. 1B). In FFPE cases with low DLRSpread values (<0.3), there was an excellent equivalence in aberration calls, with nearly identical results for frozen samples (92.3%–95.2%). The FFPE cases with higher DRLSpread values (>0.3) showed lower but still significant equivalence, with values ranging from 62% to 80%. Discrepancies in the aberration calls were mainly observed in CNAs smaller than 50 kb (50%) but became very infrequent in CNAs larger than 100 kb (Fig. 1C).
Finally, the main limitation associated with the use of FFPE specimens was the inability to identify the precise location of the CNA breakpoints, with low concordance values ranging from 17.1% to 38.1% in assays with good DLRSpread values and from 3.8% to 20% in poor quality assays (Fig. 1D).
Characterization of CNA in PCNSL
All PCNSL cases were characterized by highly complex genomes, with a median of 23 CNAs per patient (total of 210 CNAs; range, 17–47). Deletions were more common than gains, comprising almost 70% of the CNAs (61.5% of monoallelic and 7.1% of biallelic deletions). The remaining 30% were 1 copy gain, with the exception of 1 CNA characterized by the acquisition of 2 extra copies.
Overall, 20 chromosomal regions were recurrently affected in 3 or more cases, 9 of those being copy number losses and the remaining 11 comprising copy number gains. Deletion of 9p21.3 and gain of 12q12-q24.33 were the most common CNAs, observed in 5 of 7 cases each (Fig. 2 and Table 2), as previously found in other reports (21, 22). The minimal deleted region (MDR) at 9p21.3 targets the CDKN2A locus, being biallelically affected in 2 of 5 cases. These deletions were further confirmed using FISH. The minimal amplified region (MAR) on chromosome 12 includes almost the whole q arm (95.2 Mb). Chromosome arms 6p and 6q were also deleted in 5 cases each, but there was not a unique deleted region common to all cases. In 6p arm, several CNAs were found in cytoband 6p21 affecting HLA genes and in 2 cases the deletion was biallelic. In cases with 6q deletions, cytobands q12-q14.3, q16.3-q22.2, q22.31, and 6q25.1 were lost in 4 of 5 cases each (Fig. 3). PTPRK, previously analyzed by Cady and colleagues (11), was found deleted only in 3 patients by aCGH.
Overview of CNAs identified in this study. Bars at the left of the chromosomes represent losses, and bars at the right represent gains. The amplitude of the bars denotes the frequency within which each region was affected.
Overview of 6q status. The 3 most commonly deleted regions (MDR1–MDR3) are highlighted. None of them includes PTPRK (solid line).
Summary of the most commonly involved regions in CNA
Other frequent alterations were gains of 5p15.33-q23.3, 7p14.2-p22.3, 12p13.31, 16p12.3-p13.3, 18p11.31, 19q13.43, 21q22.11-q22.3, Xq22.1, and Xq28 and losses of 3p21.1 and 3q26.32 (Table 2). On cytoband 3q26.32, a recurrent focal deletion, was found in 3 cases encompassing one gene, the transducin (beta)-like 1 X-linked receptor 1 (TBL1XR1).
A total of 15 regions were biallelically deleted, affecting 18 genes (Table 3). Besides the aforementioned losses on 6p21 and 9p21, a recurrent biallelic deletion was found in 6q14.1, including the sole gene TMEM30A. Other genes of interest targeted by biallelic deletions were thymocyte selection-associated high-mobility group box (TOX) and the ETV6 (TEL) on 8q12.1 and 12p13, respectively. On the other hand, 2 extra copies of 7p22.1-p22.3 were found in 1 case, targeting CARD11 and another 42 genes (Table 3).
Summary of homozygous deletions and high copy gains found in this study
Finally, 144 genes were located in chromosomal breakpoints (Supplementary Table S1). Several of these genes, such as ETV6, NCOA2, and FOXP1, were previously identified as being part of fusion protein in other hematologic diseases.
Discussion
In PCNSL, the source of biological samples is often limited to FFPE specimens. Only recently, aCGH has become a feasible methodology to be used in FFPE specimens (15–17). In this study, we present the first data obtained from FFPE specimens of PCNSL cases. Our analysis confirmed the good to fair quality and reliability of the data generated from a very limited amount of FFPE tissue. The results have encouraged us to analyze larger archival tissue collections to complement the still fragmented knowledge we have of the genetics basis of PCNSL.
On the basis of our earlier observation that deletion of chromosomal region 6q22-23 by FISH correlated with shorter survival of PCNSL patients (14), one of the main goals of this study was to study this chromosomal region of interest in more detail by aCGH. Early studies in PCNSL showed loss of 1 or 2 copies of 6q22-23 in 66% of cases in a small series (23). More precise mapping implicated the PTPRK gene within the 140 kb common minimally deleted region in these cases, and deletion of this region was often associated with loss of expression of PTPRK. Furthermore, this small series also showed that LOH at the PTPRK locus, as well as lack of PTPRK protein expression, was associated with a poorer prognosis (24). Of note, 6q22 was only deleted in 3 of 5 cases with 6q losses. Our aCGH findings are in agreement with previous studies in DLBCL and other B-cell malignancies showing multiple affected regions on 6q without a unique MDR to all cases (25, 26).
Previous comprehensive copy number analysis in PCNSL was done with spatial resolutions ranging from 200 kb to several megabases (21, 22, 27). In this study, we improved the resolution of the analysis, reaching an average of 6 and 17 kb by using Agilent G3 and 244A platforms, respectively. As a result, we were able to refine previously identified MDR/MAR and identify several novel abnormalities affecting tumor suppressor genes and oncogenes potentially implicated in the pathogenesis of PCNSL.
Although including few cases, our preliminary analysis identified potential hits to be followed in larger analysis. Genetic and epigenetic alterations affecting CDKN2A (9p21.3) have been previously identified in PCNSL (21, 22, 28). Here we found CNA in 5 of 7 cases, with biallelic deletions observed in 2 cases. The resolution of the approach used provides a better appreciation of the precise prevalence of abnormalities affecting this and other critical genes.
A recurrent focal deletion was found on 3q26.32 targeting exclusively TBL1XR1. This gene is a transcriptional regulator that interacts with the corepressors of nuclear hormone receptor (NHR; ref. 29). Monoallelic deletions and the significant associated underexpression of TBL1XR1 have been recently reported in 12% to 15% of ETV6-RUNX1–positive acute lymphoblastic leukemias (30, 31). It has been hypothesized that loss of TBL1XR1 would compromise the ability of corepressor complexes to inhibit receptor activity, leading to the activation of receptor target genes in the presence of TBL1XR1 deletions (31). In fact, experiments on knocking down the expression of TBL1XR1 have removed the capacity of retinoic acid to induce gene expression (32). Of interest, TBL1XR1 is widely expressed in hematopoietic tissues and may have a key regulatory role in the nuclear factor κB (NF-κB) pathway (32) and Wnt-mediated transcription (33), thus suggesting its potential biological role in PCNSL pathogenesis.
Recurrent loss of 8q12.1, including biallelic loss in 1 case, has allowed us to refine previously described MDR (22) and to identify TOX as the target gene. TOX has been associated with CD4 T-lineage development (34). Furthermore, a reduction of the spleen IgG B-cell population in a TOX-deficient mouse may be suggestive of the TOX involvement in the B-cell differentiation arrest (34).
Additional focal monoallelic deletions affecting negative regulators of the NF-κB signaling pathway (MAP4K1, TANK, TAX1BP1, TRIB3) and cell-cycle (RB1), and immune-cell regulation (SIRPB1, CBLB, NFATC2) were also identified. Further analyses are needed to study in more detail these genes and their potential involvement in PCNSL pathogenesis.
In summary, the study reported here expanded the spectrum of chromosomal regions of interest, identified several highly prevalent regions that are thought to be biologically important, and showed that FFPE-based aCGH was feasible and reliable in PCNSL cases, thus expanding the repertoire of investigative tools in this tumor. A better understanding of the underlying mechanisms leading to PCNSL development could result in the identification of prognostic markers and therapeutic targets.
Disclosure of Potential Conflicts of Interest
R. Fonseca received research sponsorship from Cylene, Onyx, and Celgene and a patent for prognostication of MM based on genetic characterization of the disease. R. Fonseca also receives consulting fees from Medtronic, Otsuka, Celgene, Genzyme, BMS, and AMGEN.
Grant Support
The work reported here was supported in part by “Steve's Run”; the Mayo SPORE in Brain Cancer (P50CA108961; PI: B.P. O'Neill), including SPORE Supplement P50CA108961-03S1 (to B.P. O'Neill); and the Iowa/Mayo Clinic Lymphoma SPORE (P50CA97274; PI: G. Weiner).
The costs of publication of this article were defrayed in part by the payment of page charges. This article must therefore be hereby marked advertisement in accordance with 18 U.S.C. Section 1734 solely to indicate this fact.
Footnotes
Note: Supplementary data for this article are available at Clinical Cancer Research Online (http://clincancerres.aacrjournals.org/).
The Mayo SPORE in Brain Cancer (B.P. O'Neill and C. Giannini) and the Iowa/Mayo Lymphoma SPORE (E.R. McPhail, W. Macon, and A. Dogan).
- Received February 11, 2011.
- Revision received April 27, 2011.
- Accepted May 3, 2011.
- ©2011 American Association for Cancer Research.