Prevalent involvement of illegitimate V(D)J recombination in chromosome 9p21 deletions in lymphoid leukemia.

To understand molecular pathways underlying 9p21 deletions, which lead to inactivation of the p16/CDKN2A, p14/ARF, and/or p15/CDKN2B genes, in lymphoid leukemia, 30 breakpoints were cloned from 15 lymphoid leukemia cell lines. Seventeen (57%) breakpoints were mapped at five breakpoint cluster sites, BCS-LL1 to LL5, each of <15 bp. Two breakpoint cluster sites were located within the ARF and CDKN2B loci, respectively, whereas the remaining three were located >100 kb distal to the CDKN2A, ARF, and CDKN2B loci. The sequences of breakpoint junctions indicated that deletions in the 11 (73%) cell lines were mediated by illegitimate V(D)J recombination targeted at the five BCS-LL and six other sites, which contain sequences similar to recombination signal sequences for V(D)J recombination. An extrachromosomal V(D)J recombination assay indicated that BCS-LL3, at which the largest number of breakpoints (i.e. five breakpoints) was clustered, has a V(D)J recombination potential 150-fold less than the consensus recombination signal sequence. Three other BCS-LLs tested also showed V(D)J recombination potential, although it was lower than that of BCS-LL3. These results indicated that illegitimate V(D)J recombination, which was targeted at several ectopic recombination signal sequences widely distributed in 9p21, caused a large fraction of 9p21 deletions in lymphoid leukemia.

To understand molecular pathways underlying 9p21 deletions, which lead to inactivation of the p16/CDKN2A, p14/ARF, and/or p15/CDKN2B genes, in lymphoid leukemia, 30 breakpoints were cloned from 15 lymphoid leukemia cell lines. Seventeen (57%) breakpoints were mapped at five breakpoint cluster sites, BCS-LL1 to LL5, each of <15 bp. Two breakpoint cluster sites were located within the ARF and CDKN2B loci, respectively, whereas the remaining three were located >100 kb distal to the CDKN2A, ARF, and CDKN2B loci. The sequences of breakpoint junctions indicated that deletions in the 11 (73%) cell lines were mediated by illegitimate V(D)J recombination targeted at the five BCS-LL and six other sites, which contain sequences similar to recombination signal sequences for V(D)J recombination. An extrachromosomal V(D)J recombination assay indicated that BCS-LL3, at which the largest number of breakpoints (i.e. five breakpoints) was clustered, has a V(D)J recombination potential 150-fold less than the consensus recombination signal sequence. Three other BCS-LLs tested also showed V(D)J recombination potential, although it was lower than that of BCS-LL3. These results indicated that illegitimate V(D)J recombination, which was targeted at several ectopic recombination signal sequences widely distributed in 9p21, caused a large fraction of 9p21 deletions in lymphoid leukemia.
V(D)J recombination is a physiological recombination of DNAs, which occurs at the Ig and T cell receptor loci during the differentiation of lymphoid cells (1,2). These loci have recombination signal sequences (RSSs) 1 consisting of a highly con-served heptamer sequence (consensus 5Ј-CACAGTG) and an AT-rich nonamer sequence (consensus 5Ј-ACAAAAACC). The recombination-activating gene (RAG) complex generates DNA double strand breaks (DSBs) at pairs of RSSs in which one signal has a 12-bp spacer between the heptamer and nonamer (12-signal) and the other signal has a 23-bp spacer (23-signal) (1,2). This is known as the 12/23 rule. Several nucleotides are inserted as junctional additions at the junctions, and the broken DNA ends are rejoined by the nonhomologous end joining pathway (1)(2)(3). The results of V(D)J recombination are that coding junction products and, in some cases such as the Ig locus, signal junction products are left in the human genomic DNA (1). It has been suggested that recurrent chromosomal aberrations in lymphoid malignancies, such as translocations and inversions involving the Ig and T cell receptor loci, are caused by illegitimate actions of the V(D)J recombination machinery at "ectopic" RSSs, the sequences homologous to the consensus RSS at non-Ig, non-T cell receptor loci (4 -9). These chromosomal aberrations cause the activation or fusion of several oncogenes leading to leukemogenesis. In a large portion of these aberrations, insertions of nucleotides were detected between the two joined germ line sequences. In addition, recently, several DNA fragments containing the ectopic RSSs were shown to have potential to undergo V(D)J recombination in vivo (10 -12). Thus, it was indicated that illegitimate V(D)J recombination is a major cause for the occurrence of these chromosomal aberrations.
Deletion of the chromosome 9p21 region is a crucial event for the development of lymphoid leukemia irrespective of lineages, since homozygous deletions of this region have been found in 20 -50% of lymphoid leukemia both of T-and B-lineages (13,14). The chromosome 9p21 segment contains the 40-kb region encoding the p16/CDKN2A tumor suppressor gene and two other related genes, p14/ARF and p15/CDKN2B, all of which encode critical factors for the regulation of cell cycle and/or apoptosis (15,16). Previously, Cayuela et al. (17) characterized three sites of breakpoint clustering in T cell acute lymphoblastic leukemias (T-ALLs) in the 40-kb region. The three sites consisted of a major breakpoint cluster site, MTS1bcr␤, and two minor sites, MTS1bcr␣ and MTS2bcr1 (see Fig. 1A). All three sites contained heptamer-RSS-like sequences; therefore, it was suggested that illegitimate V(D)J recombination is involved in the occurrence of 9p21 deletions in T-ALL. However, in that study, structures of only a small number of breakpointcontaining regions in leukemia cells only of T-lineage were analyzed. In addition, the involvement of V(D)J recombination was suggested only by the sequences of breakpoints and was not functionally assessed. Therefore, it still remains largely obscure whether illegitimate V(D)J recombination is a major cause for 9p21 deletions in lymphoid leukemia.
In the present study, 30 breakpoints for 9p21 deletions were cloned in 15 human lymphoid leukemia cell lines consisting of nine T-and six B-lineage lines. The breakpoints were distributed at 18 different sites in the 9p21 segment. Seventeen (57%) of the breakpoints were mapped at five breakpoint cluster sites, BCS-LL1 to LL5, each of Ͻ15-bp. One of the breakpoint cluster sites, BCS-LL2, was the same as MTS1bcr␤ (17), whereas the remaining four sites were novel. Two breakpoint clustering sites were located within the 40-kb region encoding the CDKN2A, ARF, and CDKN2B genes, whereas the remaining three were located Ͼ100 kb apart from the 40-kb region. Thus, the results further supported the previous finding of breakpoint clustering for 9p21 deletions in lymphoid leukemia. All five BCS-LLs and seven other breakpoint sites contained the CAC trinucleotide, which is identical to the first three nucleotides of heptamer-RSSs indispensable for V(D)J recombination. The sequences of breakpoint junctions indicated that deletions in the majority (11 lines, 73%) of the cell lines were mediated by illegitimate V(D)J recombination preferentially targeted at BCS-LLs, and such a recombination occurred in both the T-and B-lineage cell lines. Thus, we further performed an extrachromosomal V(D)J recombination assay to functionally assess the V(D)J recombination potential of BCS-LLs. All of the four BCS-LL fragments tested were shown to have the potential to undergo V(D)J recombination as RSSs, whereas the efficiency and specificity of the potential were different among the fragments. The present results strongly indicated that illegitimate V(D)J recombination, which was targeted at RSS-like sequences widely distributed in the 9p21 segment, causes a major (Ͼ70%) fraction of 9p21 deletions in lymphoid leukemia of both T-and B-lineages. There was no evidence of V(D)J recombination at the breakpoint junctions of the remaining four (27%) cell lines, raising the possibility that 9p21 deletions in a small subset of lymphoid leukemia were caused by mechanisms other than illegitimate V(D)J recombination.
A human pre-B ALL cell line, Reh, and a human erythroleukemia cell line, HEL, were used for the extrachromosomal V(D)J recombination assay. They were maintained in RPMI 1640 medium supplemented with 10% fetal calf serum, 100 units of penicillin G/ml, 100 g of streptomycin sulfate/ml. For the Reh cell line, 50 M mercaptoethanol was also added, as described previously (11).
Cloning of Genomic DNA Fragments Containing Breakpoint Junctions-Breakpoints were cloned by the inverse PCR method (23) or genomic PCR using primers that recognize sequences flanking distal and proximal breakpoints, respectively. The inverse PCR method had some modifications as follows; briefly, DNA (1 g) was digested with 10 units of restriction enzyme with 4-or 6-base pair recognition sites. For circularization, the digested DNA was ligated in a 100-l reaction with 6 units of T4 DNA ligase (Promega, Madison, WI) and a ligation buffer at 16°C. After purification by using a QIA quick spin PCR purification kit (Qiagen, Tokyo, Japan), circular products were amplified by PCR using primers that recognize sequences located on the known part of the rearranged fragment. PCR products were sequenced directly or after they were subcloned into the pGEM-T easy vector by using a pGEM-T easy vector system I kit (Promega). To determine DNA sequences, a Big Dye Terminator cycle sequencing ready reaction kit (PerkinElmer Life Sciences) and the ABI 310 DNA Sequence System were used. Sequences of the DNA fragments obtained by the inverse PCR method were confirmed by direct sequencing of the PCR products, which were obtained by genomic PCR using the primers that recognize sequences flanking distal and proximal breakpoints, respectively.
Construction and Analysis of a Contiguous Sequence at 9p21-Sequences deposited in GenBank TM , AC000047, AC000048, AC000049, AL359922, and AL449423, were processed by the SEQUENCHER program, version 3.1.1 (GENE CODES, Ann Arbor, MI). A sequence of 291,432 bp with a single gap, whose size is unknown, at position 238,144 -238,145 was obtained. A genomic fragment of ϳ800 bp containing the gap was amplified by PCR from the BAC clone DNA, 436N8, which contains the CDKN2A, CDKN2B, and ARF loci (22). The PCR products were subcloned into the pGEM-T easy vector by using a pGEM-T easy vector system I kit (Promega) and sequenced by using a Big Dye Terminator cycle sequencing ready reaction kit (PerkinElmer Life Sciences) and the ABI 310 DNA Sequence System. The length of the gapped region was 64 bp. Consequently, a contiguous sequence of 291,496 bp, covering the MTAP, CDKN2A, CDKN2B, and ARF loci, was obtained. Sequence homology was examined by the BLAST2 program (available on the World Wide Web at www.ncbi.nlm.nih.gov/ blast/bl2seq/bl2.html).
Extrachromosomal V(D)J Recombination Assay-The human leukemia cell lines were grown logarithmically, transfected with the appropriate plasmid substrates using the DEAE-dextran electroporation method (25), and cultured for 48 h at 37°C. Subsequently, plasmid DNA was recovered from the cells and digested with DpnI. DpnI cleaves plasmid that does not lose its prokaryotic dam methylation pattern by replicating in eukaryotic cells. Because only molecules replicated in eukaryotic cells remain undigested by DpnI and will transform the bacteria, after digestion, they were electrotransformed into DH10B Escherichia coli (Invitrogen). Bacterial transformants were plated on media containing ampicillin alone (100 g/ml) and on ampicillin and chloramphenicol (100 and 33 g/ml, respectively). In bacteria, all substrates confer ampicillin resistance (designated as Amp r ). Recombinants, which resulted in the deletion of the prokaryotic transcription terminator on the substrates, allow cat (chloramphenicol acetyltransferase) gene expression in bacteria and thus confer both ampicillin and chloramphenicol resistance (designated as Amp r CP r ) (24). The ratio of Amp r CP r colonies to Amp r colonies reflects the fraction of recovered substrates that underwent recombination leading to deletions of the prokaryotic transcription terminator on the substrates. The averaging of multiple different eukaryotic transfections of the same substrate was performed by dividing the sum of the numbers of Amp r CP r colonies from each transfection by the sum of the numbers of Amp r colonies. Thus, recombination frequency was expressed as the weighted average (equal to (⌺Amp r CP r )/(⌺Amp r )).
Recombinant Analyses-Individual Amp r CP r colonies were picked up and were subjected to PCR amplification using primers 5Ј-AGGCC-CTTTCGTCTTCAAGA-3Ј and 5Ј-AACGGTGGTATATCCAGTGA-3Ј, which flank the SalI and BamHI sites in pGG49, respectively. PCR was performed under the following conditions: 10 mM Tris-HCl, 50 mM KCl, 1.5 mM MgCl 2 , 250 M dNTP each, 0.5 units of Taq polymerase (TaKaRa, Tokyo, Japan) in a volume of 20 l. PCR amplification was performed for 35 cycles consisting of denaturation at 95°C for 60 s, annealing at 55°C for 60 s, and extension at 72°C for 60 s. PCR products were purified by a QIA quick spin PCR purification kit (Qiagen, Tokyo, Japan) and were directly sequenced by a Big Dye Terminater cycle sequencing kit and the ABI 310 DNA Sequence System (PerkinElmer Life Sciences).

Cloning of Breakpoint Junctions in Lymphoid Leukemia-
Forty-two lymphoid leukemia cell lines (21 T-lineage and 21 B-lineage) were examined for homozygous deletions of the 9p21 region using 16 STS markers mapped in the 9p21 region. The markers consisted of 12 markers derived from exons of the CDKN2A, CDKN2B, ARF, and MTAP (methylthioadenosine phosphorylase) genes, two microsatellite markers (D9S1748 and D9S1752), and two novel STS markers (AC47 and C86) and were distributed in the 240-kb region in the 9p21 segment (STS in Fig. 1). Homozygous deletions at one or more of the tested loci were detected in 31 cell lines (74%). Ten (32%) of them showed deletion of all the 16 markers, indicating that the whole 240-kb region was deleted in these cell lines. Thus, they were not further investigated in this study. The remaining 21 (68%) cell lines showed retention of one or more markers among the 16 markers (Fig. 1). In 12 of the 21 cell lines, some markers from both the telomeric and centromeric regions were retained; therefore, interstitial deletions occurred in the 240-kb region. Thus, both the distal (telomeric side) and proximal (centromeric side) breakpoints were suggested to be present in the 240-kb region (cases 1-12 in Fig. 1). In the remaining nine cell lines, some of the 16 markers either from the telomeric or centromeric regions were retained; therefore, either the distal or proximal breakpoints were suggested to be present in the 240-kb region (cases 13-21 in Fig. 1). In total, 33 breakpoints, which consisted of 17 distal and 16 proximal ones, were indicated to be present in the 240-kb region (Fig. 1).
The distributions of these 33 breakpoints were further examined. The 240-kb region in 9p21 was divided into 24 regions of 10 kb, as shown in Fig. 1, and the breakpoints were mapped within the regions of "A" to "X" by using a number of STS markers prepared by the authors. Thirty breakpoints were mapped into some of the 24 regions, whereas one and two breakpoints were mapped to the A-B and I-K regions of 20 and 30 kb, respectively, since they were located in contiguous repetitive sequences extending from the A to B and I to K regions (Fig. 1). The C, T, and U regions contained 5, 10, and 5 break- points, respectively, suggesting the clustering of breakpoints (Fig. 1). Meanwhile, other 10-kb regions contained less than five breakpoints. More precise mapping of the breakpoints in the C, T, and U regions using STSs placed at 50-bp intervals revealed that four breakpoints in region C, five in T, and all five in U were clustered within a 50-bp region in the C, T, and U regions, respectively (breakpoints with asterisks in Fig. 1). The remaining breakpoints, one in C and five in T, were not mapped within 50 bp of other breakpoints. The clustered 14 breakpoints were derived from 11 cell lines (cases in orange, Fig. 1). Therefore, genomic DNA fragments containing breakpoint junctions were cloned from the 11 cell lines by inverse PCR or genomic PCR using sets of primers recognizing the regions flanking the breakpoints. We also cloned DNA fragments containing breakpoint junctions from four other cell lines, KAWAI, HPB-NULL, KOPN-30, and THP-7, as representatives for the cell lines with unclustered breakpoints (cases in green, Fig. 1). In total, breakpoints were cloned from 15 cell lines consisting of nine T-and six B-lineage ones.
The nucleotide sequences of these DNA fragments were compared with that of the 9p21 segment. All of the deletions in the 15 cell lines subjected to breakpoint cloning were simple interstitial ones (Fig. 1); thus, the structures of 30 breakpoints at 15 junctions were defined. Distal breakpoints for two cell lines, KOPN30 and PEER, were first mapped to the K and T regions (cases 9 and 12, Fig. 1), respectively; however, the breakpoints that had been joined with proximal breakpoints by inverse PCR analysis were mapped to the regions more telomeric than deduced (the J and D regions, respectively) (Fig. 1). Therefore, it was indicated that these two cell lines have two different alleles with deletions overlapping each other. Twenty-nine (97%) breakpoints consisting of 14 proximal and 15 distal breakpoints were assigned to the 250-kb region (the 240-kb segment analyzed plus an adjacent distal 10-kb segment) containing the CDKN2A, CDKN2B, ARF, and MTAP genes, whereas the proximal breakpoint of THP-7 was mapped to the region ϳ6.5 Mb proximal to the C86 locus outside the 250-kb region (Fig. 1).
Presence of Multiple Breakpoint Cluster Sites in 9p21-Four and five breakpoints mapped within a 50-bp region in the C and U regions, respectively, were clustered within a 10-and 14-bp interval, respectively. We designated the breakpoint cluster sites in C and U as BCS-LL (breakpoint cluster sites in lymphoid leukemia) 1 and 3, respectively (Fig. 2, A and B). Five breakpoints in the T region were clustered within a CA/GT tract in the D9S1748 locus. This locus was the same as MTS1bcr␤, which had been defined as a major breakpoint cluster site for primary T-ALLs in a previous study (17). The length of the CA/GT tract in the D9S1748 locus is known to be highly polymorphic, and the corresponding noncancerous DNAs for these five cell lines were not available. Therefore, these five breakpoints were assigned between the TACTGTG heptamer and the T-G/G-T dinucleotides 3Ј to the TACTGTG heptamer so that this heptamer sequence flanked the breakpoints (Fig. 2B). We designated this locus as BCS-LL2 to simplify the names of breakpoint cluster sites in this study. BCS-LL1 was located in the intron 4 of MTAP, a neighboring gene of CDKN2A and ARF. BCS-LL2 was located ϳ300 bp downstream of exon 1␤ of the ARF gene, while BCS-LL3 was located in intron 1 of the CDKN2B gene.
In addition, the presence of two other breakpoint cluster sites was indicated. Distal breakpoints of two cell lines, KOPN-K and SKW-3, were assigned to the same position of ϳ10 kb distal to the MTAP-exon 1 locus, indicating the presence of a breakpoint cluster site (Fig. 2, A and B). The distal breakpoint of PEER was located 3 bp proximal to the distal breakpoints of two primary T-ALLs, which had been previously assigned (17), indicating the presence of another site (Fig. 2, A  and B). This site was located in intron 4 of the MTAP gene. We designated these breakpoint cluster sites as BCS-LL4 and BCS-LL5, respectively.
In total, 17 (57%) of the 30 breakpoints identified in this study were located at the five BCS-LLs (Fig. 2B), while the remaining 13 breakpoints were dispersed at 13 sites not being located within 50 bp of other breakpoints. Therefore, a total of 18 sites (5 BCS-LLs plus 13 other sites) were defined as containing breakpoints.
Sequence Analysis of the Breakpoint Junction Sites-To assess the involvement of illegitimate V(D)J recombination in 9p21 deletions, RSS-like sequences were examined at these 18 sites. For this purpose, the 23-bp sequence flanking the breakpoints was searched for the 5Ј-CAC trinucleotide in the forward direction and the 3Ј-GTG trinucleotide in the reverse direction. This was due to the fact that the CAC trinucleotide in the heptamer-RSSs is the only part of the RSSs indispensable for V(D)J recombination (26,27) and nucleotide loss as a result of V(D)J recombination does not extend over 20 bp (28). Either the CAC or GTG trinucleotide was detected in total in 12 (67%) of the 18 sites (Fig. 1). Notably, the trinucleotide was detected at all of the BCS-LL sites (Fig. 2B).
In 11 (73%) cell lines consisting of seven T-and four B-lineage lines, the CAC trinucleotide flanked both the distal and proximal breakpoints (Fig. 3A). In addition, all of these cell lines had insertions of 1-10-bp nucleotides at the breakpoint junctions (Fig. 3A). Thus, they had features of V(D)J recombination (1, 2). In six of the cell lines, both the proximal and distal breakpoints were located at BCS-LLs (cases in red, Fig. 3A), whereas, in four others, either a proximal or distal breakpoint was located at BCS-LLs (cases in blue, Fig. 3A). Neither of the breakpoints were located at BCS-LLs in the remaining one cell line,  In four (27%) cell lines consisting of two T-and two B-lineage lines, evidence for V(D)J recombination was not observed at the breakpoint junctions. In two cell lines, P12/Ichikawa and KAWAI, the GTG trinucleotide flanked the proximal breakpoints, whereas the CAC trinucleotide did not flank the distal breakpoints (Fig. 3B). In one of them, P12/Ichikawa, the proximal breakpoint was located at BCS-LL2. In the remaining two cell lines, HPB-NULL and THP-7, neither the CAC nor GTG trinucleotide flanked the breakpoints. No nucleotides were inserted at the junctions of these four cell lines. Overlaps of 1-bp nucleotide were observed at the junctions of HPB-NULL and THP-7.
V(D)J Recombination Potential of BCS-LL3 Fragment-Detection of the CAC trinucleotide in the vicinity of breakpoints, especially of those at BCS-LLs, suggested that illegitimate V(D)J recombination is a major cause of 9p21 deletions. However, a variety of tetranucleotides followed CAC (Figs. 2B and 3A), and there were no obvious nonamer-RSS-like sequences 12 or 23 bp downstream of the heptamer-RSS-like sequences, CACNNNN. Therefore, it was not clear that these breakpoint junctions were generated by the V(D)J recombination machinery. Thus, we went on to do functional assessment of V(D)J recombination potential of DNA fragments containing breakpoints. For this purpose, we carried out the extrachromosomal V(D)J recombination assay using the pGG49 plasmid (24,29). pGG49 has 12-and 23-signals, which contains the consensus heptamer-and nonamer-RSSs with 12-and 23-bp intervals, in its SalI and BamHI sites, respectively (Table I). Since V(D)J recombination gives rise to rearranged plasmids with the signal joint configuration in which the 12-and 23-signals are joined with or without insertions of several nucleotides (i.e. N-nucleotides), the resultant recombinants are resistant to chloramphenicol due to the lack of the transcription terminator sequence of 247 bp (Fig. 4). The plasmid was introduced into a human pre-B ALL cell line, Reh, which has been shown to have V(D)J recombination activity (29), was harvested after 48 h incubation, and was introduced into E. coli for genetic discrimination of recombinants. The recombination frequency of pGG49 was 2.2 ϫ 10 Ϫ1 (Table II), similar to the results of previous studies (24,29). The signal joint configuration in the recombined plasmids was examined by PCR amplification of DNA fragments encompassing the 12-and 23-signals and subsequent sequencing of them. All of the 20 recombinants randomly picked and sequenced were signal joint products, since the consensus 12-and 23-signal sequences were joined with or without nucleotide insertions (Table III). pGG49 was also introduced into HEL, a leukemia cell line of the myeloid lineage, which has been shown to have no V(D)J recombination activity (29). As expected, pGG49 did not yield recombinant products, and its recombination frequency was estimated as being Ͻ5.1 ϫ 10 Ϫ5 (data not shown).
Next, we assessed the V(D)J recombination potential of the DNA fragments containing breakpoints by examining the frequency and fidelity of V(D)J recombination of modified pGG49 plasmids, in which the 12-and 23-signals were replaced with 33-and 43-bp sequences containing putative RSS sequences at the breakpoints, respectively (Table I, Fig. 4). First, we tested the BCS-LL3 fragment, in which the largest portion of breakpoints was clustered ( Fig. 2A). We constructed two plasmids, pLL3-RSS and pRSS-LL3, in which the 12-and 23-signals of pGG49 were replaced with 33-and 43-bp sequences of BCS-LL3, respectively (Table I, Fig. 4). The pLL3-RSS plasmid yielded recombinants at a frequency of 1.5 ϫ 10 Ϫ3 in Reh cells, whereas the pRSS-LL3 did not yield recombinants; thus, its recombination frequency was estimated as being at least 1000fold lower than that of pLL3-RSS (Table II). In all of the 20 recombinants from pLL3-RSS, which were randomly picked and sequenced, the BCS-LL3 and consensus 23-signal sequences were joined with or without nucleotide insertions and thus were considered to be signal joint products (Table III). This result indicated that BCS-LL3 has a specific recombination potential as a 12-signal, and its potential is 150-fold lower than that of the consensus 12-signal (Table II).
The pLL3-RSS plasmid recovered from HEL cells did not yield recombinant colonies, and its recombination frequency in HEL was estimated as being Ͻ1.6 ϫ 10 Ϫ6 (data not shown). Thus, the recombination frequency of the pLL3-RSS plasmid in Reh cells was obviously higher than that in HEL cells, indicating that the rearrangements of the pLL3-RSS plasmid in Reh cells were carried out by V(D)J recombination. To further confirm that the BCS-LL3 fragment functions as an RSS, we assessed the recombination frequency of the pLL3M-RSS plasmid in Reh cells. pLL3M-RSS had a mutant BCS-LL3 fragment positioned as a 12-signal, in which the first position of the CAC trinucleotide in a putative heptamer-RSS was substituted by a guanine (Table I). pLL3M-RSS showed a recombination frequency 30-fold lower than that of pLL3-RSS (Table II). In addition, none of the 20 recombinants were considered to be true signal joint products, since none of them retained the signal joint configuration. Instead, they had mutations or interstitial deletions in the transcription terminator or suffered gross rearrangements involving mutated BCS-LL3 and/or 23signal sequences. Therefore, the mutation introduced in the BCS-LL3 fragment significantly inhibited the production of recombinants with signal joint configuration. This result was consistent with a previous finding that the CAC trinucleotide in the heptamer-RSS is crucial for V(D)J recombination (25,26). Thus, it was further confirmed that the BCS-LL3 fragment functioned as an RSS in vivo. V(D)J Recombination Potential of BCS-LL1, -2, and -5 Fragments-Next, we assessed the recombination potentials of three other BCS-LL fragments, BCS-LL1, BCS-LL2, and BCS-LL5. The pRSS-LL1 plasmid carrying the BCS-LL1 fragment as a 23-signal yielded recombinants at a frequency of 2.4 ϫ 10 Ϫ4 , whereas the pLL1-RSS plasmid carrying the BCS-LL1 fragment as a 12-signal did not yield recombinants, and its frequency was estimated as Ͻ9.8 ϫ 10 Ϫ6 . The pLL5-RSS plasmid carrying the BCS-LL5 fragment as a 12-signal yielded recombinants at a frequency of 1.8 ϫ 10 Ϫ4 , whereas the pRSS-LL5 plasmid carrying the BCS LL5 fragment as a 23-signal did  not yield recombinants, and its frequency was estimated as Ͻ5.4 ϫ 10 Ϫ6 . Twenty recombinants from pRSS-LL1 and pLL5-RSS plasmids were picked and sequenced, respectively. All of them were considered to be signal joint products (Table III). Thus, it was indicated that the BCS-LL1 and BCS-LL5 fragments have specific recombination potential as a 23-and as a 12-signal, respectively. Their efficiencies were estimated as 900-and 1200-fold lower than that of the consensus 23-and 12-signals, respectively. The pRSS-LL1 and pLL5-RSS plasmids recovered from HEL cells did not yield recombinant colonies, and their recombination frequencies in HEL were estimated as being Ͻ1.3 ϫ 10 Ϫ6 and Ͻ4.6 ϫ 10 Ϫ6 , respectively (data not shown).
In contrast to the results above, both the pLL2-RSS and pRSS-LL2 plasmids, in which the BCS-LL2 fragment was positioned as a 12-and as a 23-signal, respectively, yielded recombinant products at similar frequencies (3.5 ϫ 10 Ϫ5 and 2.9 ϫ 10 Ϫ5 , respectively) ( Table II). All of the 18 recombinants obtained from the two experiments using the pLL2-RSS plasmid were sequenced. Sixteen of them (89%) were signal joint products (Tables II and III). However, the remaining 2 were not considered as recombinants by V(D)J recombination using the BCS-LL2 sequence as a RSS. One recombinant did not have the deletion of the transcription terminator but had two G to A substitutions in the transcription terminator sequence. The other recombinant apparently retained a wild type sequence of the terminator, suggesting that unknown genetic alterations on the substrate caused the transcriptional activation of the cat gene. Therefore, we estimated the V(D)J recombination frequency of pLL2-RSS as 3.1 ϫ 10 Ϫ5 (equal to 3.5 ϫ 10 Ϫ5 ϫ 16/18) (Table II). pLL2-RSS recovered from HEL cells did not yield recombinant colonies, and their recombination frequencies in HEL were estimated as being Ͻ7.0 ϫ 10 Ϫ6 (data not shown). Thus, the V(D)J recombination frequency of BCS-LL2 as a 12-signal was higher in Reh than in HEL, supporting the fact that the fragment functioned as a 12-signal in vivo. Twenty recombinants from pRSS-LL2 were also randomly picked and sequenced. Four of them (20%) were signal joint products (Tables II and III), and they consisted of one identical recombinant with the same 5-bp insertion. The remaining 16 were not considered as V(D)J recombinants using the BCS-LL2 sequence as a RSS. Fifteen of them consisted of six types of recombinants in which the transcription terminator was deleted along with the BCS-LL2 or the 12-signal sequence. The remaining one allowed amplification of multiple DNA fragments, suggesting that gross rearrangement occurred in this clone. Thus, we estimated the V(D)J recombination frequency of pRSS-LL2 as 5.9 ϫ 10 Ϫ6 (equal to 2.9 ϫ 10 Ϫ5 ϫ 4/20) (Table  II). pRSS-LL2 plasmids recovered from HEL cells did not yield recombinant colonies, and the recombination frequency was Ͻ2.9 ϫ 10 Ϫ5 (data not shown). Thus, it was inferred that the V(D)J recombination potential of BCS-LL2 as a 23-signal was present but at a low level.
V(D)J Recombination by Pairing BCS-LL3 and BCS-LL1-An additional plasmid, pLL3-LL1, in which BCS-LL3 and BCS-LL1 were inserted in the place of the 12-and the 23-signals, respectively, was tested to examine whether or not V(D)J recombination by pairing the BCS-LL3 and BCS-LL1 fragments occurs. If we assume that the BCS-LL3 and BCS-LL1 fragments undertake recombination at efficiencies 1 ⁄150 and 1 ⁄900 less than consensus RSSs, respectively (Table II), the recombination frequency of pLL3-LL1 could be calculated as being 1.6 ϫ 10 Ϫ6 (equal to 2.2 ϫ 10 Ϫ1 ϫ 1/150 ϫ 1/900). The pLL3-LL1 plasmid yielded two recombinant clones in the six experiments (Table II). However, sequence analysis of these two recombinants indicated that they were not the signal joint products generated by pairing BCS-LL3 and BCS-LL1 but rather recombinants with different interstitial deletions within the transcription terminator. Hence, the actual recombination frequency was Ͻ0.9 ϫ 10 Ϫ6 ; thus, the use of two inefficient RSSs did not give rise to V(D)J recombination with enough efficiency to be detected by the assay used.

DISCUSSION
In the present study, the distribution of 33 breakpoints for 9p21 deletions in 21 lymphoid leukemia were examined in the 240-kb segment covering the 40-kb region containing the CDKN2A, ARF, and CDKN2B loci, the region where the breakpoint distribution was previously examined in T-ALL (17). This breakpoint mapping and the subsequent breakpoint cloning enabled us to identify three breakpoint cluster sites, BCS-LL1 to LL3. These three sites contained four or five breakpoints; thus, they were considered as major breakpoint cluster sites (Fig. 2, A and B). BCS-LL2 was the same as MTS1bcr␤ in the ARF locus, where the significant portion (Ͼ60%) of the breakpoints in T-ALL had been previously mapped; therefore, the result of Cayuela et al. (17) was confirmed. In the present study, an additional major breakpoint cluster site, BCS-LL3, was identified in the CDKN2B locus, and another site, BCS-LL1, was identified in the MTAP locus of Ͼ100 kb distal to the CDKN2A, ARF, and CDKN2B loci ( Fig. 2A). Furthermore, two additional breakpoint cluster sites, BCS-LL4 and -LL5, were also identified at loci Ͼ100 kb distal to these loci ( Fig. 2A). Since two or three breakpoints among those identified by us or Cayuela et al. (17) were located at these sites, they were considered as being minor breakpoint cluster sites. Cayuela et al. (17) also identified two minor breakpoint cluster sites, MTS1bcr␣ and MTS2bcr1, in the vicinity of the CDKN2A and CDKN2B loci, respectively ( Fig. 2A). However, none of the breakpoints in this study were mapped at these two sites. Therefore, they are likely to be minor cluster sites as previously reported (17). Thus, the present result further supported the previous finding that breakpoints for 9p21 deletions in lymphoid leukemia are clustered. In addition, it was found that both the breakpoints of T-and B-lineage cell lines were located at the BCS-LL sites. Thus, the present result also indicates that the breakpoint cluster sites are susceptible to DNA breaks during lymphoid leukemogenesis irrespective of cell lineages.
All BCS-LLs contained the CAC trinucleotide, which is identical to the first three nucleotides of heptamer-RSSs indispensable for V(D)J recombination. In addition, 7 of the 13 sites, where a single breakpoint was mapped in this study, also contained the CAC trinucleotide within 23 bp of each breakpoint. Therefore, in total, 12 (67%) of the 18 sites contained the CAC trinucleotide (Fig. 1). Since the probability of finding CAC in the genome is 1 in 64 bp ( 1 ⁄4 ϫ 1 ⁄4 ϫ 1 ⁄4 ϭ 1 ⁄64), 18 sites of 23-bp, each of which contains 21 trinucleotides, could contain six copies of CAC (18 ϫ 21 ϫ 1 ⁄64 ϭ 5.9). Therefore, the CAC trinucleotide was overrepresented at the breakpoints. Breakpoint junctions in 11 (73.3%) cell lines consisting of seven T-and four B-lineage lines showed typical evidence of V(D)J recombination (Fig. 3A). The CAC trinucleotide, which flanked both the distal and proximal breakpoints, was deleted with the segment encoding the CDKN2A, ARF, and CDKN2B genes, and nucleotides of 1-10 bp were inserted at the breakpoint junctions. The deletions in these cases were typical of coding junctions of V(D)J recombination. DSBs generated by the RAG complex are processed to the P-nucleotide and/or N-nucleotide additions in the case of coding junctions (1). Three of the 22 breakpoints in these 11 cases, distal ones of L-SMY and CCRF-CEM and a proximal one of KOPT-K1, were located next to the CAC trinucleotide, respectively, and one or two nucleotides complementary to the germ line sequences upstream of the CAC trinucleotide were inserted at the junctions (underlined in Fig. 3A). These nucleotides are likely to be P-nucleotides. The remaining 19 breakpoints were located 1-15 bp away from the heptamer-RSS-like sequences; therefore, inserted N-nucleotides (rather than P-nucleotides) probably account for nucleotide sequences beside those breakpoints. Thus, overrepresentation of CAC in the vicinity of breakpoints as well as the sequences of the breakpoint junctions indicated that the majority (Ͼ70%) of 9p21 deletions in lymphoid leukemia of both T-and B-lineages are caused by illegitimate V(D)J recombination.
In the remaining four (27%) cell lines, V(D)J recombination was not evident at the breakpoint junctions (Fig. 3B). In two T-lineage cell lines, P12/Ichikawa and KAWAI, only proximal breakpoints were flanked by the CAC trinucleotide, and nucleotides were not inserted at the junctions. In two B-lineage cell lines, HPB-NULL and THP-7, breakpoints were not flanked by the CAC trinucleotide. Therefore, it was likely that deletions in these cell lines were generated by mechanisms other than illegitimate V(D)J recombination. Joining of nonhomologous DNA without evidence of V(D)J recombination has been detected at the breakpoint junctions for the interstitial deletions in nonhematological malignancies, including 9p21 deletions in two cases of glioma (30 -33). Therefore, as hypothesized in nonhematological malignancies (3,34,35), 9p21 deletions in a subset of lymphoid leukemias, such as these four cell lines, may have also been generated by nonhomologous end joining of DSBs caused by unknown mechanisms. Meanwhile, the proximal breakpoint of P12/Ichikawa was located at BCS-LL2. Therefore, it might be that the DSB at the proximal breakpoint in this cell line was generated by the illegitimate action of the RAG complex. The broken DNA ends generated by the RAG  Notably, in 10 of the 11 cell lines whose 9p21 deletions were possibly caused by illegitimate V(D)J recombination, one or both of the proximal and distal breakpoints were located at BCS-LLs. However, the RSS-like sequences, even those at BCS-LLs, showed only a partial homology to the consensus RSSs (Fig. 2B, Table I). Thus, we assessed the V(D)J recombination potential of DNA fragments containing the RSS-like sequences. We subjected four BCS-LL fragments, BCS-LL1, -LL2, -LL3, and -LL5, to the extrachromosomal V(D)J recombination assay in Reh cells with V(D)J recombination activity (29). The BCS-LL3 fragment positioned as a 12-signal (i.e. pLL3-RSS) showed the highest recombination frequency (Table  II), indicating that this fragment had the highest recombination potential among the four fragments tested. This result was consistent with the fact that the largest portions (i.e. five breakpoints) of the breakpoints were clustered at BCS-LL3 ( Fig. 2A). Other cluster sites, BCS-LL1 and BCS-LL5, that were located Ͼ100-kb apart from the CDKN2A, ARF, and CDKN2B loci, were also shown to undergo V(D)J recombination as RSSs, respectively. It was inferred that BCS-LL2 has a V(D)J recombination potential both as 12-and 23-signals, but its potential, especially as a 23-signal, was considerably lower than other BCS-LL fragments. This result was contrary to the fact that a considerable number of breakpoints were clustered at the BCS-LL2 site in this (Fig. 2B) and a previous studies (17). BCS-LL2 is located in a CA/GT tract, and this tract was not cloned into the pLL2-RSS and pRSS-LL2 plasmids (Table I). CA/GT tracts are speculated to adopt the Z-DNA structure (17), which might enhance V(D)J recombination; thus, the recombination potential of BCS-LL2 could be enhanced by the CA/GT tract.
In the present study, all four BCS-LL fragments tested showed potential as RSSs, although the efficiency and specificity as RSS were different among the fragments. Especially, the estimated V(D)J recombination potential of BCS-LL1 and -LL3, two major cluster sites, were similar to that of other ectopic RSSs, which have been predicted to pair with physiological RSSs to give rise to recurrent translocations in lymphoid leukemia (11). In the 9p21 deletions analyzed in this study, the BCS-LL3 site was joined with the BCS-LL1 site, but not with the BCS-LL5 site, to give rise to deletions (Fig. 3A). This is consistent with the fact that the BCS-LL3 and BCS-LL5 sites preferentially acted as 12-signals, whereas the BCS-LL1 site acted as a 23-signal. The BCS-LL5 site with a 12-signal potential was joined with the BCS-LL2 site, for which a weak potential as a 23-signal was suggested (Fig. 3A). Thus, it is likely that the occurrence of 9p21 deletions conformed to the 12/23 rule (1, 2). V(D)J recombination by pairing of two ectopic RSSs, BCS-LL1 and BCS-LL3, was not recapitulated in the extrachromosomal recombination assay; however, this was conceivable, since its predicted recombination frequency was too low to be detected by this system (11). Thus, the present results strongly support the fact that illegitimate V(D)J recombination, targeted at ectopic RSSs in 9p21, causes a large fraction of 9p21 deletions in lymphoid leukemia.
The majority of the breakpoints were clustered at BCS-LLs despite the fact that the 9p21 segment analyzed in this study contained thousands of CAC trinucleotides. Thus, it is possible that BCS-LLs have specific structures enhancing V(D)J recombination other than the CAC trinucleotide. It is noted that BCS-LL1, -LL3, and -LL5 specifically functioned as a 12-or 23-signal contrary to the fact that they have no obvious nonamer-RSS-like sequences (Table I). In addition, the recombination potential of BCS-LLs was not correlated to the degrees of their homologies with the consensus RSSs. These results indicate that BCS-LL fragments can take specific configurations as RSSs, irrespective of their homologies with the consensus RSSs. In addition, it is also noted that all of the five BCS-LLs were located in, or upstream of, the genes in this region ( Fig. 2A). The MTAP and CDKN2A genes are expressed ubiquitously in a variety of human tissues, including hematopoietic progenitor cells (20,36,37). Recent studies have indicated that the open chromatin structure associated with gene expression is a critical factor determining the efficiency of V(D)J recombination, probably changing the accessibility of the V(D)J recombination machinery (38,39). Therefore, the local chromatin structure of BCS-LLs might further enhance the susceptibility to illegitimate V(D)J recombination. Illegitimate V(D)J recombination has been indicated as being involved in a variety of chromosomal aberrations in lymphoid leukemias (8,9). However, only a few candidate ectopic RSSs have been functionally assessed for their potential as RSSs (10 -12). Thus, genomic structures underlying illegitimate V(D)J recombination should be further investigated to define the mechanisms underlying 9p21 deletions and other recurrent chromosomal aberrations that occur in lymphoid leukemogenesis.