Genome-Wide Array-Based Comparative Genomic Hybridization of Diffuse Large B-Cell Lymphoma

INTRODUCTION

Diffuse large B-cell lymphoma (DLBCL) comprises some 30% of non-Hodgkin’s lymphoma cases and is clinically heterogeneous (1 , 2) . Recent microarray analyses of transcripts of DLBCL specimen have clearly shown biologically distinct subtypes in DLBCL that are also clinically relevant (3 , 4) . Although several genetic alterations have been identified as etiologically associated with DLBCL (5 , 6) , genome-wide screening has been insufficient. The recently developed array-based comparative genomic hybridization (array CGH) techniques allow high-throughput analysis of copy number changes of a genome at high resolution and accuracy throughout the whole genome. The quantitative measurement of DNA copy number thus obtained may facilitate identification of tumor-related genes (7, 8, 9) . We recently established our own array CGH using a glass slide on which 2088 bacterial artificial chromosome/P-1 phage-derived artificial chromosome (BAC/PAC) DNA were spotted in duplicate with an average resolution of 1.5 Mb (10) . This enabled us to identify a novel gene, C13orf25, as overexpressed in B-cell lymphoma cell lines and DLBCL patients with 13q31-q32 amplifications (10) . These results warrant application of array CGH analysis to DLBCL genome widely.

CD5 is a cell surface molecule physiologically expressed on T cells and on subsets of B cells residing in the mantle zone of lymphoid organs and the peritoneal cavity. Clinically, CD5 expression is often associated with chronic lymphocytic leukemia (11) and mantle cell lymphoma (MCL; ref. 12 ). For DLBCL cases, we have identified CD5 expression as a marker of poor prognosis, whereas CD5-positive (CD5⁺) DLBCL is associated with elderly onset, female predominance, and the frequent involvement of extranodal sites (13) . Microarray analyses of transcripts showing different expressions by CD5⁺ and CD5-negative (CD5⁻) DLBCL cases also indicate that CD5⁺ DLBCL is a distinct disease entity (14 , 15) .

In this study, we used the array CGH technique to analyze 26 CD5⁺ and 44 CD5⁻ DLBCL cases and identified genomic aberrations shared by both groups, as well as those specific to CD5⁺ DLBCL. Our results may provide both insights into the genomic basis of lymphoma genesis and genetic markers useful for the identification of clinically distinct subgroups of DLBCL.

MATERIALS AND METHODS

Patients.

We analyzed DNA samples of 26 cases of de novo CD5⁺ DLBCL and 44 cases of CD5⁻ DLBCL. These samples were obtained with informed consent from patients at Aichi Cancer Center and collaborating institutions under the approval of the Institutional Review Boards. All patients were reported previously (13 , 16 , 17) . The median age was 61 years and 56 years for the CD5⁺ and CD5⁻ cases, respectively. Among the CD5⁺ cases, 68% were female, 80% were at advanced stages (III to IV), 72% had elevated lactate dehydrogenase, 24% had a poor performance status, and 28% had extranodal site(s) of involvement. In the CD5⁻ cases, 41% were female, 52% had advanced stages (III to IV), 45% had elevated lactate dehydrogenase, 10% had a poor performance status, and 38% had extranodal site(s) of involvement. All of the samples were obtained from tumors at diagnosis before any treatment was given.

DNA Samples.

DNA was extracted using a standard phenol chloroform method from lymphoma specimens from a total of 70 DLBCL cases: 26 cases of CD5⁺ DLBCL and 44 cases of CD5⁻ DLBCL. Normal DNA was prepared from peripheral-blood lymphocytes of healthy male donors.

Malignant Lymphoma Cell Line.

Cell line used in this study was OCI-LY13.2 (DLBCL, kindly provided by Dr. Ricardo Dalla-Favera of Columbia University, New York, NY; ref. 18 ). OCI-LY13.2 was maintained in RPMI 1640 supplemented with 10% fetal bovine serum at 37°C in 5% CO₂ −95% air.

Selection of BAC/PAC Clones (BAC/PACs) for Array CGH.

The array consisted of 2088 BAC/PACs, covering whole human genome with ∼1.5 Mb of resolution. BACs were derived from RP11 and RP13 libraries, and PACs were derived from RP1, RP3, RP4, and RP5 libraries. BAC/PACs used were selected based on information from National Center for Biotechnology Information and Ensembl Genome Data Resources. 7 8 These clones were obtained from the BACPAC Resource Center at the Children’s Hospital (Oakland Research Institute, Oakland, CA). 9 Clones were ordered from chromosomes 1 to 22 and X. Within each chromosome, clones were ordered from chromosomes 1 to 22 and X on the basis of Ensembl Genome Data Resources of Sanger Center Institute, January 2004 version. All of the clones used for array CGH were confirmed for their location on chromosomes by fluorescence in situ hybridization. Clone names and their locations on chromosomes are available on request.

DNA Amplification for Spotting on Slides.

Ten ng of BAC (or PAC) DNA were used as the template for degenerate oligonucleotide-primed PCR (19) with the 5′-amine-modified degenerate oligonucleotide-primed primer (5′-CCGACTCGAGNNNNNNATGTGG-3′ where n = A, C, G, or T) and amplified on a TaKaRa PCR Thermal Cycler MP (TaKaRa, Tokyo, Japan) using Ex Taq polymerase (TaKaRa). A 3-minute, 94°C denaturation step was followed by 25 cycles of 94°C for 30 seconds, a 37°C to 72°C linear ramp for 10 minutes, and 72°C for 1 minute, with a final 7-minute extension at 72°C.

DNA Spotting and Quality Control of Glass Slides.

Degenerate oligonucleotide-primed PCR products were ethanol precipitated and dissolved in distilled water, and then an equal volume of DNA spotting solution DSP0050 (Matsunami, Osaka, Japan) was added (1 μg/μL). The resulting DNA samples were robotically spotted by an inkjet technique (NGK, Nagoya, Japan) in duplicate onto CodeLink-activated slides (Amersham Biosciences, Piscataway, NJ). In this study, we used only glass slides on which it had been confirmed that all 2088 clones had been spotted completely and uniformly in duplicate.

Array Hybridization.

The array fabrication and hybridization was performed according to the method described by Pinkel et al. (20) and Hodgson et al. (21) . One μg of tested (tumor or normal) and of referenced (normal) DNA was digested with DpnII and labeled with the BioPrime DNA labeling system (Invitrogen-Life Technologies, Inc., Tokyo, Japan) using Cy3-dUTP and Cy5-dUTP (Amersham Pharmacia Biotech, Piscataway, NJ) for the tested and referenced DNA, respectively. Unincorporated fluorescent nucleotides were removed by means of Sephadex G-50 spin columns (Amersham Biosciences). Tested and referenced DNA were mixed with 50 μg of human Cot-1 DNA (Invitrogen-Life Technologies, Inc.), precipitated, and resuspended in 45 μL of hybridization mixture consisting of 50% formamide, 10% dextran sulfate, 2× SSC, 4% SDS, and 10 μg/μL yeast tRNA (Invitrogen-Life Technologies, Inc.). The hybridization solution was heated to 73°C for 5 minutes to denature the DNA and then incubated for 45 minutes at 37°C to block repetitive sequences. The glass slides spotted with DNA were denatured in 70% formamide/2× SSC at 73°C for 4 minutes, then dehydrated in cold 70, 85, and 100% ethanol for 5 minutes each and air-dried. Hybridization was performed for 66 to 72 hours in a container on a slowly rocking table with 200 μL of 50% formamide/2× SSC, followed by posthybridization washings in 50% formamide/2× SSC for 15 minutes at 50°C, in 2× SSC/0.1% SDS for 30 minutes at 50°C, and in PN buffer (0.1 mol/L NaH₂PO₄ and 0.1 mol/L Na₂HPO₄ to attain pH 8 and 0.1% NP40) for 15 minutes at room temperature. The glass slides were then rinsed in 2× SSC at room temperature and finally dehydrated in 70, 85, and 100% ethanol at room temperature for 2 minutes each and air-dried. The slides were scanned with an Agilent Micro Array Scanner (Agilent Technologies, Palo Alto, CA), and the acquired array images were analyzed with Genepix Pro 4.1 (Axon Instruments, Inc., Foster City, CA). After automatic segmentation of the DNA spots and subtraction of the local background, intensities of the signals were determined. Subsequently, ratios of the signal intensity of two dyes (Cy3 intensity/Cy5 intensity) were calculated for each spot, converted into log₂ ratios on an Excel sheet in the order of chromosomal positions, and then normalized. For centering the log₂ ratio of normal components at 0, we normalized the log₂ ratio of each of the samples according to the following method. The medium log₂ ratio value for all clones was computed, and the clones were selected with a log₂ ratio more than “the median + SD × A” or less than “the median − SD × A.” “A” was visually defined as the normal region by referring to the log₂ ratio plots of all clones for each experiment and was also assigned an approximate range from 0.5 to 1.0. We then computed the mean log₂ ratio values for the selected clones and designated these value X. Finally, we obtained the Y value by subtracting X from the log₂ ratio for each clone. In this study, each of the log₂ ratios was analyzed on the basis of the Y value.

For the array, six simultaneous hybridizations of normal male versus normal male were performed to define the normal variation for the log₂ ratio. A total of 122 clones with <10% of the mean fluorescence intensity of all of the clones, with the most extreme average test over reference ratio deviations from 1.0 and with the largest SD in this set of normal controls, was excluded from additional analyses. Thus, we analyzed a total of 1966 clones (covered 2988 Mb, 1.5 Mb of resolution) for additional analysis. A total of 1907 clones (covered 2834 Mb) of 1966 was from 1ptel to 22qtel. Because >96% of the measured fluorescence log₂ ratio values of each spot (2 × 1966 clones) ranged from +0.2 to −0.2, the thresholds for the log₂ ratio of gains and losses were set at the log₂ ratio of +0.2 and −0.2, respectively. Linearity between copy number and values of log₂ ratio was confirmed by the use of human fibroblast cell lines (22) that had different copy number of X chromosomes as described previously (10) .

Fluorescence In situ Hybridization Analysis.

Metaphase and interphase chromosomes were prepared from normal male lymphocytes and OCI LY13.2 cell line. Dual color fluorescence in situ hybridization analysis was conducted as described previously (10 , 23) .

Statistical Analysis.

To analyze genomic regions for statistically significant differences between the two patient groups, the data set was constructed by defining genomic alterations as copy number gains for log₂ ratio thresholds of +0.2 or more and as copy number loss for thresholds of −0.2 or less. Clones showing a gain (log₂ ratio > +0.2) were inputted as 1 versus no-gain clones (log₂ ratio ≤ +0.2) as 0 on an Excel template for each case. Similarly, loss clones (log₂ ratio < −0.2) were inputted as 1 versus no-loss clones (log₂ ratio ≥ −0.2) as 0 on another Excel template for each case. Cases showing genomic gain or loss were counted with Excel for each single clone (1966 clones in total) in the CD5⁺ group or CD5⁻ group. Data analyses were then carried out for the following purposes: (a) comparison of frequencies of gain or loss of each single clone between the CD5⁺ and CD5⁻ groups (1966 tests each for gain and loss, 3932 tests in total); (b) comparison of overall survival between cases showing gain or loss of a single clone and cases without respective gain or loss (1966 tests for each gain and loss with or without CD5 expression, in total 7864 tests maximum). Fisher’s exact test for probability was used for the former comparison, and a log-rank test for comparing survival curves of the two groups was used for the latter. P for screening of candidate clones for an analysis was <0.05. When a candidate clone was identified, the clone’s continuity with the subsequent clones was examined. In cases where the nth clone and succeeding k clones (k ≥ 0) were found to be candidate clones, the P value for continual association was calculated as: on the assumption that each clone is independent throughout the entire genome. Because multiple tests (11,796 tests maximum) were used, the conventional Bonferroni procedure was applied to define the α-error for the final conclusion. Therefore, we defined a value for the calculation given above as < 0.05/12,000 (= 4.2 × 10^-6) as statistically significant (24) . All of the statistical analyses were conducted with a statistical package STATA, version 8 (College Station, TX).

RESULTS

Genomic Profiles and Data Analysis for DLBCL Cases.

Array CGH analysis was performed to compare genomic alterations in CD5⁺ and CD5⁻ DLBCL cases. All of the clones on chromosome X (57 clones) were separately analyzed because of sex mismatching. Of the 70 DLBCL cases enrolled, 4 cases (1 of CD5⁺ DLBCL and 3 of CD5⁻ DLBCL) did not show any genomic aberrations. The remaining 25 cases of CD5⁺ DLBCL and 41 cases of CD5⁻ DLBCL were then subjected to the data analysis. Fig. 1 ⇓ shows the entire genomic profiles of two representative CD5⁺ samples (Fig. 1, A and B) ⇓ and one CD5⁻ sample (Fig. 1C) ⇓ . Copy number changes were easily detectable at a high-resolution genome widely. Regions of high-level gain/amplification (defined as log₂ ratio > +1.0), and regions suggestive of homozygous loss/deletion (defined as log₂ ratio < −1.0) were also easily detected, as were regions showing low-level gain/amplification (defined as log₂ ratio +0.2 to +1.0), as well as regions suggestive of heterozygous loss/deletion (defined as log₂ ratio −1.0 to −0.2).

The entire tumor set comprised an average copy number gain of 311.1 Mb or 10.9% of the genome (6.8 regions) and an average copy number loss of 174.4 Mb or 6.8% of the genome (5.9 regions). The CD5⁺ group showed a larger average fraction of copy number gain (370.9 Mb, 13.4%) than the CD5⁻ group (311.1Mb, 10.9%), whereas the former (110.5 Mb, 3.8%) contained a smaller average fraction of copy number loss than the latter (174.4 Mb, 6.8%). The average number of total alterations consisted of 8.1 regions of gain and 5.4 regions of loss in the CD5⁺ group and 6.0 regions of gain and 6.4 regions of loss in the CD5⁻ group.

We defined region of gain or loss as (a) continuously ordered three clones showing gain or loss or (b) clones showing high copy number gain (log₂ ratio > +1.0) or homozygous loss (log₂ ratio < −1.0; e.g., a gain of 2p15, a loss of 3p14.2 and a loss of 9p21). Recurrent region was defined as a region seen in >20% of cases. The most frequently gained or lost BAC/PACs in each of the recurrent regions are listed in Tables 1 ⇓ and 2 ⇓ .

In the CD5⁺ group (25 cases), recurrent regions of gain were 1q21.2-q32.3, 1q42.2-q42.3, 2p15-p16.1, 3, 5p13.2-p13.1, 6p22.3-p25.3, 7p22.2-q31.1, 8q24.13-q24.22, 11q22.1-q25, 12, 13q21.1-q34, 16p13.3-q21, 17q23.2-q24.3, 18, 19p13.13-q13.43, and X and recurrent regions of loss were 1p36.23-p36.32, 1p36.13-p36.21, 1p34.3-p35.1, 1q43-q44, 3p14.2, 6q14.1-q27, 8p23.3-p21.2, 9p21, 15q13.1-q14, and 17p11.2-p13.3.

In the CD5⁻ group (41 cases), recurrent regions of gain were 1q21.2-q31.1, 1q32.1-q32.2, 3p25.2-q29, 5p13.1-p13.2, 5p13.2-p14.1, 6p12.3-p25.3, 7, 8q24.13-q24.21, 9p24.2-p13.2, 11q23.2-q24.2, 12q13.2-q21.2, 16p13.3, 18, and X and recurrent regions of loss were 1p36.23-p36.32, 3p14.2, 6q12-q25.2, 6q27, 9p21, 15q15.2-q21.1, and 17p11.2-p13.3.

Regions of gain observed in >20% of cases in both the CD5⁺ and CD5⁻ groups were 1q21.2-q31.1, 1q32.1-q32.2, 3p25.2-q29, 5p13.1-p13.2, 6p21.1-p25.3, 7p22.2-q31.1, 8q24.13-q24.21, 11q23.2-q24.3, 12q13.2-q21.2, 16p13.3, 18, and X and regions of loss were 1p36.31-p36.32, 3p14.2, 6q14.1-q25.2, 6q27, 9p21, and 17p11.2-p13.3.

Among the recurrent regions, the gain of 2p15 (BAC, RP11-17D23), loss of 3p14.2 (BAC, RP11-48E21), and loss of 9p21.1 (BAC, RP11-149I2) showed genomic aberrations by single or two continuous clones. Three of 13 cases with gain of 2p15 showed high-level copy number gain (log₂ ratio > +1.0) at a restricted position of the genome (∼2 Mb resolution level) encompassing two continuously ordered BACs, RP11-17D23 (500 kb centromeric to the REL gene) and RP11-511I11 (1.8 Mb centromeric to BAC, RP11-17D23).

Eighteen of the 66 cases (28%) showed loss of 3p14.2. Among the clones contained in 3p14.2, BAC, RP11-48E21 (including FHIT) only was lost in 13 of the 18 cases (7 cases of CD5⁺ and 11 cases of CD5⁻ DLBCL), with no surrounding BACs showing any obvious copy number losses (Fig. 2A) ⇓ . Among the 13 cases with a single loss, 2 cases showed homozygous loss at BAC, RP11-48E21 (log₂ ratio < −1.0). 3p14.2 was also deleted in the cell line OCI-LY13.2 established from a patient with aggressive malignant lymphoma (18) . Fluorescence in situ hybridization analyses results were consistent with those of the array CGH (Fig. 2, B and C) ⇓ . Similarly, the BAC clone RP11-149I2, including INK4a/ARF, only was lost in 9 of the 32 cases that showed loss of 9p21 among a total of 66 DLBCL cases, and 2 of the 9 cases showed homozygous loss.

Finally, we analyzed X chromosomes for male patients only (10 cases in the CD5⁺ group and 25 cases in the CD5⁻ group). Two CD5⁺ cases and three CD5⁻ cases showed low-grade copy number gains throughout the entire X chromosome but no high-grade amplification. Heterozygous losses were found at Xq21 in two cases in the CD5⁺ group, whereas no homozygous loss region was found at chromosome X.

Genomic Copy Number Changes Characteristic of CD5⁺ DLBCL.

We next compared the frequency of gain and loss of clones in the CD5⁺ and CD5⁻ groups. Screening on a single-clone basis for candidate clones showed that 48 clones were more frequently (P < 0.05) gained or lost in the CD5⁺ than in the CD5⁻ group. Among these 48 clones, 6 of 10p14-p15.3, 3 of 12p12, 3 of 16p12, and 9 of 19q13.32-q13.43 were continuous clones in terms of the whole genome mapping position according to the Ensemble Genome Data Resources of the Sanger Center Institute (January 2004 version). The remaining 27 clones showed individual P < 0.05, i.e., with no neighboring clones showing such significance.

Twenty clones were identified as being lost significantly more frequently in the CD5⁺ group than in the CD5⁻ group. Among these 20 clones, 3 clones of 1q43, 6 of 1q43-q44, 2 of 8p23.3, and 5 of 8p23.1-p23.2 were continuous, and the remaining 4 clones showed individual P < 0.05 with no neighboring clones showing such significance. In the CD5⁻ group, 10 gained clones and 6 lost clones showed individual P < 0.05 with no neighboring clones showing such significance.

Clones screened on a single-clone basis as described above were subsequently subjected to multiple comparison corrections to find clones statistically relevant in terms of differences in frequency between the CD5⁺ and the CD5⁻ groups. Because values of continuous clones with gains of 10p14-p15.3 and 19q13.32-q13.43 and losses of 1q43-q44 and 8p23.1-p23.2 were <4.2 × 10^-6, these regions of gain and loss were determined to be characteristic of the CD5⁺ group. No significantly frequently occurring region of gain or loss was found in the CD5⁻ group after multiple comparison corrections had been performed. Regions characteristic of CD5⁺ DLBCL and the BACs they contained are listed in Table 3 ⇓ .

Genomic Copy Number Changes Affecting Prognosis of DLBCL.

The next step was the analysis, using the Kaplan-Meier method and log-rank test, of the probabilities of survival of the cases that had been stratified according to the presence or absence of one of the specific genomic gains or losses. All clones that showed aberrant copy number changes in either the CD5⁺ or the CD5⁻ group were subjected to screening, followed by multiple comparison corrections for the clones that showed P < 0.05 in screening with the log-rank test.

In the CD5⁺ group, 252 clones (98 gained and 154 lost clones) showed P < 0.05 for prognosis according to log-rank test results on a single-clone basis. Eighteen of the lost clones were continuous at 1p34.3-p36.21, and multiple comparison corrections showed them to have deleterious effects on survival. Forty-four of the gained clones were of 13q. Among the 44 clones, 29 at 13q21.1-q31.3 and 15 at 13q31.3-q34 were continuous and were identified by multiple comparison corrections as being linked to poor prognosis. The remaining 190 gained or lost clones fell short of statistical significance.

In the CD5⁻ group, 252 clones (197 gained and 55 lost clones) showed P < 0.05 for prognosis on a single-clone basis. Thirty-four of these clones were of 5p. Among the 34 clones, 19 clones at 5p14.2-p15.33 and 5 clones at 5p12-p13.2 were continuous, and multiple comparison corrections indicated they had favorable impacts on survival. The remaining 218 gained and lost clones fell short of statistical significance. CD5⁺ DLBCL cases with gain of 13q21.1-q34 showed significantly inferior survival than did CD5⁺ cases without such gain (Fig. 3A) ⇓ . In contrast, gain at the corresponding region did not affect survival of CD5⁻ DLBCL cases. Similarly, CD5⁺ DLBCL cases with loss of 1p34.3-p36.21 showed significantly inferior survival than did CD5⁺ cases without such loss (Fig. 3B) ⇓ , whereas loss at the corresponding region did not affect survival of CD5⁻ DLBCL cases. Conversely, CD5⁻ DLBCL cases with a gain of 5p showed superior overall survival (Fig. 3C) ⇓ to those without such gain, but a gain of 5p had no impact on survival of CD5⁺ DLBCL cases. A list of BAC/PACs associating with prognosis of CD5⁺ or CD5⁻ DLBCL and prognostic data can be found in Supplementary Data 1.

DISCUSSION

Array CGH methods have been successfully used for the high-resolution analysis of genomic alterations not only in a variety of solid tumors (7, 8, 9 , 25 , 26) but also in hematological malignancy (27, 28, 29, 30) . In this study, we used our own array CGH (10) to analyze 70 cases of DLBCL and identified regions of genomic gain and loss of DLBCLs that were relevant to clinical subtypes and patient survival.

Array CGH detected aberrations of several loci that were undetectable by chromosomal CGH (also known as conventional CGH). Loss of 3p14.2 was one such aberration. This lost locus was detected by array CGH in 18 of 66 DLBCL cases but in none by chromosomal CGH. The responsible regions for 13 of these 18 cases were covered by a single BAC, RP11-48E21, which included the FHIT tumor suppressor gene. This clearly demonstrates that array CGH is more sensitive than chromosomal CGH. These findings also suggest that array CGH is a useful tool for identifying and narrowing down to the target genes. It should be noted, however, that the efficacy of array CGH is limited to identification of copy number changes of genome, and it can therefore not detect chromosomal translocations, mutations, and epigenetic events that could affect gene expressions. The use of the array CGH method in combination with other newly developed technologies such as microarray analysis of transcripts and SKY analysis of chromosomes may further facilitate our understanding of molecular events underlying DLBCL.

We reported previously that chromosomal CGH analyses of patient samples identified six recurrent regions of gain (3q, 6p, 11q21-q24, 12q, 13q22-q32, and 18q) and four recurrent regions of loss (1p, 6q, 17p, and 19p; ref. 17 ). Array CGH analysis of the same set of patient samples found several novel recurrent regions of gain such as 1q32, 5p13, 7p22-q31, 8q24.13-q24.21, and 16p13, as well as recurrent regions of loss such as 3p14.2 and 9p21. However, loss of 19p detected by chromosomal CGH could not be confirmed in array analysis. The reasons for this apparent discrepancy between chromosomal CGH and array CGH are not yet entirely clear. One possibility is that chromosome 19 contains blocks of heterochromatin that are difficult to assess with chromosomal CGH.

It is therefore possible that the discrepancy is due to the unreliable results obtained with chromosomal CGH for these regions. Other researchers reported that the use of chromosomal CGH methods resulted in the identification of 1q21-q23, 2p12-p16, 3q26-q27, 7q11, 8q24, 9q34, 11cen-q23, 12p, 12cen-q13, 13q32, 16p12, 16q21, 18q21-q22, and 22q12 as recurrent regions of gain and 1p34-ptel, 6q23-qtel, 8p22-ptel, 17p12, and 22q as recurrent regions of loss (31, 32, 33) . Our array CGH could not identify 9q34 and 22q12 as recurrent regions of gain. Gains of 9q34 and 22q12 were found in only 5 cases (7.6%) and 6 cases (9.1%), respectively, of a total of 66 DLBCL cases. This discrepancy could be due to differences in the patient samples analyzed and in the sensitivity of the methods used.

Array CGH analysis has been applied by Martinez-Climent et al. (29) to follicular lymphoma and DLBCL transformed from follicular lymphoma. They reported that 14 regions of gain and 9 regions of loss were acquired as a result of the transformation. Among these 23 regions, gains of 4p12-pter, 9q13-q31, and 17q21 and losses of 4q21-q23, 5q21-q23, 9q31-qter, 11q24-q25, and 13q14-q21 were not found as recurrent genomic aberrations in our study of de novo DLBCL. Conversely, we found genomic alterations such as gains of 11q24, 13q13, and 16p13 and loss of 3p14 in our study of de novo DLBCL but not in the reported cases of transformed DLBCL. These findings suggest de novo DLBCL and transformed DLBCL are different in terms of the genes responsible for lymphomagenesis.

Array CGH analysis identified largely identical genomic aberration patterns in the CD5⁺ and in CD5⁻ groups. However, gains of 10p14-p15.3 and 19q13.32–13.43 and losses of 1q43-q44 and 8p23.1-p23.2 were found to be characteristic of CD5⁺ DLBCL. These findings, in addition to characteristic clinical behavior, indicate that CD5⁺ DLBCL is a distinct entity. It has not been demonstrated yet that genes included in the regions of 10p14-p15.3 and 1q43-q44 are linked to malignancy, but the 19q13.2-q13.43 region includes tumor-related genes such as BAX, PEG3 (34) , CD37, and IL4R1, with the last one having been identified as a gene responsible for primary mediastinal diffuse large B-cell lymphoma (35) . Genomic loss of 8p23 has frequently been found in leukemic MCL (36) , and this deletion may be linked with leukemic dissemination and poor prognosis for patients with MCL. Given that, in addition to MCL, 8p23 is recurrently lost in CD5⁺ DLBCL, one could speculate that this locus may contain tumor suppressor genes accounting for poor prognosis of patients with both CD5⁺ DLBCL and leukemic MCL.

It was also found that although gain of 13q21.1-q34 and loss of 1p34.3-p36.21 had deleterious impacts on the survival of CD5⁺ cases, these regions had no such impact on CD5⁻ cases. We are currently analyzing expressions of genes that are included in the loci specifically gained or lost in CD5⁺ DLBCL to gain additional insights into their roles in lymphoma development. One possible candidate gene for 13q31-q32 amplification is C13orf25, which has been found in the case of gain of 13q31-q32 in B-cell lymphoma. C13orf25 contains several microRNAs (miR-17, miR-18, miR-19a, miR19b-1, miR-20, and miR-92-1) that might be associated with lymphomagenesis (10 , 37) . There have been no reports, however, on tumor suppressor gene(s) of 1p34.3-p36.21.

In contrast to the CD5⁺ cases, we were not able to find genomic regions, either gained or lost, which were characteristic of CD5⁻ DLBCL, although gain of 5p was found to be associated with a favorable survival. More cases need to be analyzed, however, to clarify the prognostic significance of this gain.

To summarize, we subjected DLBCL cases to array CGH analysis and found genomic regions recurrently altered in DLBCL. By comparing CD5⁺ and CD5⁻ cases, we were able to identify genomic alterations specific to the CD5⁺ DLBCL group. Array CGH analysis can thus be expected to provide new insights into the genetic background of lymphomagenesis.