Activation-Induced Cytidine Deaminase-Dependent DNA Breaks in Class Switch Recombination Occur during G1 Phase of the Cell Cycle and Depend upon Mismatch Repair1

Ab class switching occurs by an intrachromosomal recombination and requires generation of double-strand breaks (DSBs) in Ig switch (S) regions. Activation-induced cytidine deaminase (AID) converts cytosines in S regions to uracils, which are excised by uracil DNA glycosylase (UNG). Repair of the resulting abasic sites would yield single-strand breaks (SSBs), but how these SSBs are converted to DSBs is unclear. In mouse splenic B cells, we find that AID-dependent DSBs occur in Sμ mainly in the G1 phase of the cell cycle, indicating they are not created by replication across SSBs. Also, G1 phase cells express AID, UNG, and mismatch repair (MMR) proteins and possess UNG activity. We find fewer S region DSBs in MMR-deficient B cells than in wild-type B cells, and still fewer in MMR-deficient/SμTR−/− B cells, where targets for AID are sparse. These DSBs occur predominantly at AID targets. We also show that nucleotide excision repair does not contribute to class switching. Our data support the hypothesis that MMR is required to convert SSBs into DSBs when SSBs on opposite strands are too distal to form DSBs spontaneously.

A ntibody class switch recombination (CSR) 3 involves the replacement of the IgM C region gene (C) by a downstream C region gene, e.g., C␥, C␣, or C, and can improve the efficacy of the immune response. The recombination occurs via the formation of double-strand breaks (DSBs) in switch (S) region DNA located upstream of the C region genes, and the joining of two different S regions by an end-joining mechanism, resulting in deletion of the intervening DNA from the genome (1). CSR requires activation induced cytidine deaminase (AID), which converts cytosines in DNA to uracils (2)(3)(4)(5)(6)(7). The target for AID is ssDNA, which can be generated during transcription, and only transcriptionally active S regions undergo CSR (1, 4, 8 -10). S regions consist of tandem repeats (TR) that are unique to each isotype, although all contain numerous targets for AID: the hotspot motif WRC/GYW, where W ϭ A or T, R ϭ G or A, and Y ϭ C or T (6,11,12). The underlined C is determinated by AID.
Repair of the dU residues resulting from AID activity leads to DNA breaks necessary for recombination. Uracil in DNA is removed by uracil DNA glycosylase (UNG) and CSR is severely reduced in mice that lack UNG and in patients with mutations in UNG (13,14). UNG activity generates abasic sites that could be recognized and cleaved by AP endonucleases (APE) to create single-strand breaks (SSBs), and recent data indicate that APE is important for CSR (53). AID-and UNG-dependent DSBs have been detected in S regions, and the breaks occur preferentially at G:C bp in AID hot-spot motifs, which is consistent with cleavage by APE at the nucleotide deaminated by AID (14 -17). The current data support the conclusion that repair of the dU residues generates SSBs, but SSBs must be converted to DSBs to excise the DNA segment intervening between S and the downstream S region.
Currently, nothing is known about the mechanism by which SSBs in S regions become DSBs. In this study, we test three hypotheses regarding how DSBs are created during CSR. SSBs initiated by AID activity could be converted into DSBs during replication, when the elongating DNA strand reaches a nick on the template strand. However, ␥H2AX and Nbs-1, two proteins involved in CSR, are found associated with the IgH locus during G 1 /early S phase, but not during S/G 2 M phase in splenic B cells undergoing CSR (18). To address this issue, we analyzed the cell cycle regulation of DSBs in S regions, with emphasis on separation of G 1 phase from S phase cells.
Nucleotide excision repair (NER) could recognize a variety of damage intermediates resulting from AID activity and could introduce DNA breaks with its associated endonucleases Ercc1-Xeroderma pigmentosum F (XPF) and XPG (19). A minor role for Ercc1 in CSR has been described (20), but it is not known whether Ercc1-XPF acts during CSR in conjunction with the NER pathway or independently as a structure-specific endonuclease. We tested whether XPA, which is essential for NER, is involved in CSR.
Nicks sufficiently close (Ͻ5 bp apart) on opposite strands can form DSBs spontaneously, as shown by experiments in which the restriction enzyme I-Sce1 can produce a chromosomal DSB (21). However, distal SSBs will not lead to DSBs, due to stability of the DNA duplex, unless the DNA ends are processed. AID is known to deaminate dC nucleotides on both the transcribed and nontranscribed strands (22,23), and the density of AID hot spots occurring in the S region TRs could lead to nicks on opposite strands in close proximity to each other. However, deletion of the STR region results in only a modest (2-fold) reduction in CSR (24). CSR can occur upstream of, as well as within, the STR region (25,26), although recombination outside of the STRs requires the DNA mismatch repair (MMR) pathway (26,27). In fact, CSR is nearly ablated in STR Ϫ/Ϫ B cells that also lack the MMR proteins Msh2 or Mlh1 (28) (J. Eccleston, C. E. Schrader, J. Stavnezer, and E. Selsing, manuscript in preparation). Because MMR is required for recombination outside the STR region where AID hot spots are relatively infrequent, we have proposed that MMR is needed to convert SSBs into DSBs when the single-strand nicks are too far apart to form spontaneous DSBs (29). This hypothesis is consistent with the finding that CSR does not require MMR, but is reduced 2to 5-fold in MMR-deficient B cells, i.e., at least 50% of CSR events require MMR, depending on the isotype (27, 30 -34). To test this hypothesis, we analyzed S DSBs in B cells from mice deficient in MMR proteins that have the WT S region or the STR deletion. Altogether, the data support the hypothesis that distal SSBs are converted by MMR into DSBs, constituting an important step in CSR.

Materials and Methods
Mice AID-deficient mice were obtained from T. Honjo (Kyoto University, Kyoto, Japan). Msh2-deficient mice were obtained from T. Mak (University of Toronto, Toronto, Canada). Mlh1-and Pms2-deficient mice were obtained from R. M. Liskay (Oregon Health Sciences University, Portland, OR). XPA-deficient mice were obtained from H. van Steeg (National Institute of Health, Bilthoven, The Netherlands) (35). These lines have all been extensively backcrossed to C57BL/6. xpa ϩ/Ϫ mice were bred to ung ϩ/Ϫ mice (obtained from D. Barnes and T. Lindahl, London Research Institute, London, U.K.) to generate doubly heterozygous mice, the breeding of which generated wild-type (WT), xpa Ϫ/Ϫ , ung Ϫ/Ϫ , and xpa Ϫ/Ϫ ung Ϫ/Ϫ littermates used for the experiment shown in Fig. 3. STR ϩ/Ϫ mice were bred to mlh1 ϩ/Ϫ and to msh2 ϩ/Ϫ mice to generate double knockouts. WT (ϩ/ϩ) littermates from MMR-deficient mice were used as controls for all experiments (except that shown in Fig. 3) and the results were identical for all WT littermates. All mice were housed in the same room of the Institutional Animal Care and Use Committee-approved specific pathogenfree facility at University of Massachusetts Medical School and were bred and used under guidelines formulated by the University of Massachusetts Animal Care and Use Committee.

Cell cycle sorting
Splenic B cells were cultured for 2 days as described above, after which cell density was adjusted to 1 ϫ 10 6 cells/ml and stimulated B cells were incubated for 90 min at 37°C with 3.5 g/ml Hoechst 33342 (Invitrogen Life Technologies) in HBSS containing 1% FCS. Cells were finally resuspended in 1 ml of HBSS with 1% FCS; 7-aminoactinomycin D was added (0.6 g/ml) and cells were sorted by flow cytometry based on DNA content using a UV laser-equipped FACSVantage SE (BD Biosciences).

UNG assay
Whole cell extracts were prepared from splenic B cells activated for 2 days, washed in ice-cold PBS, and lysed in buffer I (10 mM Tris-HCl (pH 7.8), 200 mM KCl with Complete protease inhibitors (Roche)). An equal volume of buffer II was added (buffer I with 2 mM EDTA, 40% glycerol, 0.2% Nonidet P-40, and 2 mM DTT added) before tumbling for 1 h at 4°C. Supernatants were stored at Ϫ80°C. Nuclear extracts were made from sorted cells by lysis in buffer A (10 mM HEPES (pH 8), 10 mM KCl, 0.1

Cloning, identification, and sequence analysis of PCR products
LM-PCR products were cloned into the vector pCR4-TOPO (Invitrogen Life Technologies) and sequenced by Macrogen using T3 and T7 primers. Cloned breaks in S were aligned with germline S sequenced from C57BL/6 chromosome 12 (GenBank Accession number AC073553) with numbering starting at nt 136,645. This is the 5Ј S primer-binding site and ϳ800 nt upstream of the beginning of the TRs.

AID-dependent DSBs are predominantly detected in the G 1 phase of the cell cycle
A possible mechanism for conversion of SSBs into DSBs is DNA replication, and many of the proteins involved in CSR are associated with DNA surveillance and repair during replication, e.g., MMR proteins and UNG (37). To determine when AID-dependent DSBs are made, we examined the cell cycle regulation of S DSBs. DSBs dependent on AID and UNG are induced in the S regions of B cells activated in vitro to undergo CSR and can be detected by linker LM-PCR (15). We activated splenic B cells for 2 days with LPS and IL-4 to induce IgG1 CSR or LPS and anti-␦-dextran (a-␦-dex) to induce IgG3 CSR, stained with Hoechst 33342, and then sorted cells into G 1 and S/G 2 /M fractions on the basis of DNA content (Fig. 1A). Dead cells were excluded by 7-aminoactinomycin D staining. At 48 h, there are no undivided cells in these cultures, as determined by CFSE staining (our unpublished data). As shown in Fig. 1B, AID-dependent   DSBs are detected almost exclusively in the G 1 fraction in cells activated under either condition. There are a few breaks detected in S/G 2 /M phase, but there are about as many in AID-deficient B cells. These data are therefore inconsistent with the hypothesis that DSBs detected in S in switching B cells are due to replication across SSBs. Although a few DSBs are detected in aid Ϫ/Ϫ cells, they do not specifically occur at the G:C base pair in the AID target hot spots, unlike DSBs in WT cells (15). They might be due to mechanically induced breaks or apoptosis occurring during the procedure, or perhaps due to replication.

Proteins important for CSR are expressed in G 1 -phase cells
We then asked whether the levels of AID, UNG, APE1, and MMR proteins are regulated by the cell cycle. B cells activated with LPS and a-␦-dex were sorted into G 1 , S, and G 2/ M fractions ( Fig. 2A) and extracts were prepared for Western blotting (Fig. 2, B-D). AID protein is difficult to detect in nuclear extracts, but is expressed in the cytosol in all phases of the cell cycle (Fig. 2C). UNG is abundant both in the nucleus (Fig. 2B) and cytoplasm (Fig. 2C) of G 1 -and S-phase cells. The multiple UNG bands observed in nuclear extracts represent differently phosphorylated species (38). The Ab should also detect the mitochondrial form of UNG in the cytosol. We observed a decrease in UNG in the cytosol as the cells progress through cell cycle, consistent with a report that UNG is down-regulated in late S phase (39). APE1 and the MMR proteins Msh2, Msh6, and Mlh1 are all expressed in the nucleus throughout the cell cycle (Fig. 2D). We conclude that proteins important for CSR are present in all phases of the cell cycle, and specifically in G 1 phase when AID-dependent DSBs are detectable.

G 1 -phase cells have UNG activity
UNG is not present in resting (G 0 ) cells (14) and has been proposed to act in S phase at replication forks during CSR and somatic hypermutation (40). G 1 -phase B cells have not been examined for UNG activity. We prepared nuclear extracts from G 1 -and S-phase B cells activated for 2 days and assayed UNG activity by an oligonucleotide cleavage assay (Fig. 2E). Activity is clearly detectable in as little as 10 ng of nuclear extract from G 1 -phase cells, which has as much activity as extracts from S-phase cells. As this substrate is single stranded, the predominant activity detected here is likely to be due to UNG and not SMUG-1, as SMUG-1 has 800-fold lower activity on ssDNA than UNG (k cat /K m ) (41). From these combined results, we conclude that AID-dependent DSBs in S are made and resolved in S in G 1 phase and are not due to replication across SSBs.

The role of Ercc1 in CSR does not involve the NER pathway
We next asked whether NER might contribute to the generation of DSBs during CSR. We have previously shown a minor role for  Ercc1 in CSR (20). The heterodimer Ercc1-XPF is an essential component of the NER pathway, but Ercc1-XPF can also act as a structure-specific endonuclease in the absence of the rest of the NER pathway, cutting at the junction of ssDNA and dsDNA (19). Here, we analyzed CSR in vitro in B cells from mice lacking XPA, a protein essential for NER. We found no reduction in CSR in xpa Ϫ/Ϫ B cells relative to cells from WT littermates (Fig. 3A). In the absence of UNG, it seemed possible that the accumulation of uracils might result in distortion of the DNA helix sufficient to create a substrate for NER, so we also analyzed CSR in ung Ϫ/Ϫ xpa Ϫ/Ϫ B cells CSR is severely reduced in B cells deficient in UNG, as reported by others (13,14), but still detectable relative to AID-deficient B cells. However, CSR is not further reduced in ung Ϫ/Ϫ xpa Ϫ/Ϫ B cells relative to B cells from ung Ϫ/Ϫ littermates (Fig. 3, B and C). In this experiment, there was a reduction in IgG1 and IgG2a CSR in xpa Ϫ/Ϫ cells, but this was not typical as can be seen from the compiled data in Fig. 3A. We conclude that the minor role played by Ercc1-XPF in CSR most likely involves its ability to cut at the junction of ssDNA and dsDNA (discussed below and in Ref. 29), and does not involve other components of the NER pathway.

S-region DSBs are reduced in MMR-deficient B cells
The hypothesis that SSBs are converted to DSBs in S regions by MMR predicts fewer DSBs in S in MMR-deficient B cells relative to WT cells. To test this prediction, LM-PCR was performed on genomic DNA isolated from B cells 2 days after induction of CSR. A representative Southern blot of LM-PCR fragments amplified with a 5Ј S-specific primer from WT, msh2 Ϫ/Ϫ , mlh1 Ϫ/Ϫ , and pms2 Ϫ/Ϫ B cells is shown in Fig. 4A. Blunt DSBs in MMRdeficient cells are reduced compared with WT cells analyzed in parallel. Quantitation by densitometry scanning of the autoradiographs from several experiments demonstrates that MMR deficiency results in ϳ20 -50% of the number of S DSBs in WT cells (Fig. 4B). Thus, blunt DSBs induced in S during CSR are reduced in MMR-deficient cells to a degree consistent with the reduction observed in CSR. As described in the introduction, CSR in B cells lacking the Ig S TR region (STR Ϫ/Ϫ ) is severely reduced (90 -95%) in the absence of Msh2 or Mlh1. We used LM-PCR to determine whether this might be due to a decrease in DSBs. As shown in Fig. 5, DSBs in STR Ϫ/Ϫ B cells are detected at only a slightly lower frequency than in WT cells (65% of WT), consistent with the 2-fold reduction in CSR in these B cells (24), but when the cells also lack Msh2 or Mlh1, DSBs are reduced to only 20% of that seen in WT cells. The number of DSBs detected in Msh2/STR and Mlh1/STR double knockout cells is almost as low as in AID-deficient cells.

Sites of DSBs in STR-deficient cells
Relative to the WT S region, the STR Ϫ/Ϫ intron has a 2-fold lower incidence of AID hot spots (WRC/GYW, where the underlined nucleotide is the target site), and the strongly preferred hotspot GCT (11, 33) is 3.6-fold less frequent than in WT S (Fig. 6). We asked whether the DSBs in STR Ϫ/Ϫ cells occur preferentially at these remaining hot spots by cloning and sequencing LM-PCR products from WT and STR Ϫ/Ϫ B cells. Table I presents the percent of DSBs occurring at the G:C or A:T base pair, at all AID hot spots (GYW), and at the four different GYW motifs (GCT, GCA, GTT, and GTA), as well as the frequency of occurrence of these motifs in the S-region sequence in WT and STR Ϫ/Ϫ cells. The percentage of DSBs located at GYW sequences is increased in STR Ϫ/Ϫ B cells (51.6% compared with 40.5% in WT), supporting the importance of this sequence for AID targeting. Further analysis of the breakpoints used in the STR Ϫ/Ϫ intron support previous conclusions that GCT/AGC is the strongest AID hot spot. The preference for breaks at GCT in WT S is 3.3-fold over random (35.4% of DSBs at GCT vs 10.6% frequency of GCT in the sequence, p Ͻ 0.001). However, GCT occurs in the WT S sequence at a much higher frequency than the other GYW hot spots (Table I). In the STR Ϫ/Ϫ intron, the frequency at which GCT occurs relative to the other GYW motifs is lower; interestingly, DSBs at GCT occur 10-fold more often than predicted by the sequence (29% of DSBs are at GCT compared with 2.9% frequency in the sequence, p Ͻ 0.001). These results confirm that the preference of AID for GCT hot spots previously found in vitro with purified AID and from analyses of somatic hypermutation (6,11,12,42,43) is also true during CSR, and is even more apparent when these motifs are less abundant. We hypothesize that the reduction in AID hot spots in STR Ϫ/Ϫ B cells results in SSBs that are farther apart, and thus less likely to be sufficiently close to form a DSB spontaneously. DSB formation and CSR are therefore more dependent on MMR in these mice.

Discussion
The results from our studies support the hypothesis that distal SSBs are converted into DSBs by MMR recognition and repair of dU:dG mismatches introduced by AID. We obtained data arguing against an alternative model for conversion of SSBs to DSBs, i.e., replication across a SSB, as we detected very few DSBs in S/G 2 /M phase cells. Instead, the AID-dependent DSBs occur in the G 1 phase. These results are in agreement with previous reports showing that AID-dependent ␥H2AX/Nbs1 foci colocalize with IgH loci during the G 1 /early S phase in splenic B cells activated to switch (18). They are also consistent with the finding that switch recombination is not accompanied by sister-chromatid exchange (44), and that a specific protein complex binds near the transcriptional initiation site for germline ␥1 transcription during G 1 /early S phase, but not during G 2 /M phase in splenic B cells activated to undergo CSR (45). We also found that AID, UNG, and MMR proteins are present in the G 1 phase of the cell cycle, and that extracts from G 1 -phase B cells possess UNG activity.
MMR proteins could convert SSBs into DSBs in the course of normal MMR activity if nicks are present on both strands. Both AID and UNG prefer ssDNA substrates and appear to act on both DNA strands during transcription (1,12,22,46). We hypothesize that the DNA duplex reforms before all of the dUs can be removed by UNG, and MMR would then compete more effectively than UNG for repair of dUs in duplex DNA. AID is a highly processive enzyme and although UNG has a very high catalytic rate, it appears to be unable to remove all the dUs introduced by AID into S regions (6,12,22,23,41). UNG is likely, however, to create numerous abasic sites, which could then be nicked by APE to generate entry sites for excision by Exo1 in the 5Ј to 3Ј direction. If excision continued until a SSB on the opposite strand is reached, a DSB would be formed. A similar model has been proposed for Escherichia coli dam cells exposed to the methylating agent MNNG, in which DSBs arise when MMR excision on one strand encounters a nick generated by APE on the other (47). Depending on the orientation of the SSBs relative to each other, some DSBs created by MMR activity will be blunt and some will have 5Ј or 3Ј overhangs. 5Ј overhangs can be filled in by polymerase to become blunt, and we have proposed that Ercc1-XPF can remove 3Ј overhangs (20,29). Ercc1-XPF cuts 3Ј overhangs at the junction of ssDNA and dsDNA without requiring additional NER proteins (48). Consistent with Ercc1-XPF acting independently of NER in CSR, we found that XPA, an essential NER protein, is not involved in CSR even in the absence of UNG, where helix-distorting lesions recognizable by NER might form.
If the SSBs are the results of AID-instigated lesions, the resulting blunt DSBs would occur preferentially at the G:C base pair in AID hot spots, which is consistent with the location of DSBs in S determined by LM-PCR (Table I) (15,29). No significant differences were observed between the sites of S DSBs in WT and MMR-deficient B cells (Table II). In the absence of MMR, we predict that the only DSBs formed would do so spontaneously from closely spaced nicks. The staggered DSBs produced in the absence of MMR could then be end processed as described above to form the blunt DSBs at the G:C base pair in AID hot spots detected by LM-PCR. The staggered ends can be detected by LM-PCR by pretreatment with T4 DNA polymerase before ligation (15,17). We find a 3-fold increase in DSBs after T4 treatment of DNA from both WT and MMR-deficient cells (data not shown).
It has been reported that 94% of S-S␥3 junctions in Msh2deficient cells occur within the S TRs, which contain numerous closely spaced GAGCT sequences, and only 5% occur 5Ј to the TRs, whereas in WT cells a greater proportion (17%) of the junctions occur 5Ј to the repeats (26,27). These results predict that blunt DSBs in MMR-deficient B cells would localize preferentially to the S TRs. However, we found a similar fraction of DSBs occur 5Ј to the S TRs in WT and msh2 Ϫ/Ϫ cells, 42 and 40%, respectively. This difference from the previous reports might be due to differences in the analysis methods. We used a PCR primer located at the 5Ј side of S, and extension would terminate at the first break, favoring detection of breaks located upstream of S.
We consistently find fewer blunt DSBs in B cells from mice lacking Mlh1 or Pms2 than in those lacking Msh2. This might be explained by the recent finding that Mlh1 decreases Exo1 processivity (49). If Exo1 excises farther in the absence of Mlh1-Pms2, perhaps past the nick on the opposite strand, long single-strand tails and fewer blunt DSBs would be generated. Long single-strand overhangs might also explain the increase in microhomology found at CSR junctions from mlh1 Ϫ/Ϫ and pms2 Ϫ/Ϫ mice. In addition, it has recently been shown that the Mlh1-Pms2 heterodimer has endonuclease activity that introduces nicks on either side of the mismatch (50). This activity appears to be restricted to the previously nicked strand (50), and so by itself would not result in DSBs during CSR. Furthermore, as the Mlh1-Pms2 endonuclease does not appear to be restricted to a specific DNA sequence, and S DSBs strongly favor the G:C base pair, this activity is unlikely to contribute significantly to S-region breaks.
Our results are consistent with the model that MMR is more important for CSR in situations when SSBs are limiting, as when AID targets are scarce. Thus, loss of both MMR and the STR region results in a dramatic reduction in switching and a reduction in S DSBs. Although we cannot determine the sites of the SSBs that gave rise to any given DSB, we were able to confirm that DSBs occur preferentially at AID hot spots even when the frequency of hot spots is low. We propose that only SSBs on opposite strands within a few nucleotides of each other can form a DSB in the absence of MMR. It is possible that MMR may play a significant role in DSB formation at other loci at which AID might initiate DSB formation, such as bcl6 and c-myc (51,52). AID targets in these genes are infrequent and therefore, DSB formation, which can lead to translocation and tumor development, might be more dependent on MMR. The data are reported as a percent of the DSBs at the indicated base pair or sequence motif. b WT (ϩ/ϩ) littermates from several DNA repair mutants, all of which gave identical results.
c The frequency at which the nucleotide or sequence motif occurs in the S sequence.
d Significantly targeted relative to the DNA sequence, p Յ 0.037 (Fisher's exact t test). e The underlined G indicates the G:C base pair at which the DSB occurs within the hot spots, reading the top strand sequence.
f Total number of DSB sites analyzed. g nt 1, Nucleotide 136,645 in the chromosomal 12 sequence from C57BL/6 (GenBank Accession AC073553).