Rad52 competes with Ku70/Ku86 for binding to S-region DSB ends to modulate antibody class-switch DNA recombination

Antibody class-switch DNA recombination (CSR) is initiated by AID-introduced DSBs in the switch (S) regions targeted for recombination, as effected by Ku70/Ku86-mediated NHEJ. Ku-deficient B cells, however, undergo (reduced) CSR through an alternative(A)-NHEJ pathway, which introduces microhomologies in S–S junctions. As microhomology-mediated end-joining requires annealing of single-strand DNA ends, we addressed the contribution of single-strand annealing factors HR Rad52 and translesion DNA polymerase θ to CSR. Compared with their Rad52+/+ counterparts, which display normal CSR, Rad52−/− B cells show increased CSR, fewer intra-Sμ region recombinations, no/minimal microhomologies in S–S junctions, decreased c-Myc/IgH translocations and increased Ku70/Ku86 recruitment to S-region DSB ends. Rad52 competes with Ku70/Ku86 for binding to S-region DSB ends. It also facilitates a Ku-independent DSB repair, which favours intra-S region recombination and mediates, particularly in Ku absence, inter-S–S recombination, as emphasized by the significantly greater CSR reduction in Rad52−/− versus Rad52+/+ B cells on Ku86 knockdown.

I mmunoglobulin (Ig) class-switch DNA recombination (CSR) and somatic hypermutation (SHM) are central to the maturation of the antibody response 1-4 . CSR endows antibodies with new biological effector functions by exchanging the gene encoding the Ig heavy chain constant region (C H ) with a downstream C H region. By introducing mainly point mutations in Ig V(D)J sequences, SHM provides the structural substrate for antigen-mediated selection of higher-affinity antibody mutants 1,2,4 . Similar to SHM, CSR requires activation-induced cytidine deaminase (AID)-mediated generation of DNA lesions 1,2,4 . AID, expressed in activated B cells, deaminates deoxycytosines to yield deoxyuridine:deoxyguanine mispairs 2,5 . These mispairs trigger DNA repair processes that lead to insertion of double-strand DNA breaks (DSBs) in the upstream (donor) and downstream (acceptor) switch (S) regions (CSR) 2,6 . Synapse of a S region, such as Sm, DSB ends with DSB ends of a downstream S region, such as Sg1, leads to deletion of the intervening DNA which is released as extrachromosomal S circle, and juxtaposition of a V H DJ H exon to a downstream C H exon cluster (in the above case Cg1), thereby completing the CSR process 2,4 . Other outcomes can occur. Multiple DSBs are introduced into each of the S regions that will be the targets of recombination-Sm being particularly prone to accumulating many DSBs. DSBs in a given S region can synapse with DSBs within the same S region, thereby yielding intra-S region deletions and non-CSR events. S-region DSB ends can also recombine with the DSB ends in other chromosomes to yield translocations, including c-Myc/IgH translocations 7 .
In CSR and possibly in c-Myc/IgH inter-chromosomal translocations, A-NHEJ is initiated by the DNA damage sensor Parp1 and an early HR element, the DSB end-processing factor CtIP, which facilitates DSB resection to generate protruding ('staggered') ends 28,29 . These are annealed through stretches of complementarity 21,27 , which leads to introduction of microhomologies at S-S (and c-Myc-IgH) junctions 16,18,29 . We contend here that microhomology-mediated A-NHEJ in CSR and c-Myc/IgH translocations also critically relies on another HR factor Rad52, a DNA-binding element that promotes annealing of complementary DSB single-strand ends 8,30,31 . Rad52 plays a central role in HR DSB repair and is also involved in HR-independent DSB repair 32 . We previously showed that Rad52 is recruited together with Rad51, another HR factor, to AID-resected DSB protruding ends (Rad51 recruitment to DNA DSBs is dependent on Rad52) in the human IgH locus during antibody diversification 33 . In addition to Rad52, the translesion DNA polymerase y (Poly), which promotes annealing of complementary single DNA strands, may also be involved in Ku-independent CSR. Poly facilitates MMEJ [34][35][36] . It bypasses lesions by inserting and extending past mispairs. Poly also copies an undamaged DNA template efficiently but in an error-prone manner and, as we have shown, plays a significant role in Ig locus SHM 37 . Finally, Poly can mediate DSB synapses in chromosomal translocation and is essential for cell survival when HR is impaired 35 .
Here we addressed the role of single-strand DNA annealing factors Rad52 and Poly in CSR. Using Rad52 À / À and Poly À / À B cells in vitro together with molecular genetic methods, we found that Rad52 deficiency profoundly altered CSR to all Ig classes, whereas Poly deficiency did not. We validated the in vitro findings by analysing specific class-switched antibody responses in Rad52 À / À and Poly À / À mice. We adapted chromatin immunoprecipitation (ChIP) and competition assays involving recombinant proteins, to analyse the recruitment of Rad52 (RAD52) and Ku70/Ku86 (KU70/KU86) to CSR-targeted S-region DSB ends. We studied the expression kinetics of Rad52 (RAD52) and Poly (POLy), and compared them with that of Ku70/Ku86 (KU70/KU86) in mouse and human B cells on exposure to Aicda (AICDA)/CSR-inducing stimuli. Further, we determined the impact of Rad52 deficiency on c-Myc/IgH translocations associated with CSR in p53 À / À B cells-the p53 tumour suppressor is essential for protecting B cells from c-Myc/IgH translocations and p53 deficiency significantly enhances AID-dependent c-Myc/IgH translocations without detectable effect on CSR 38 . Finally, we enforced Rad52 expression in normal B cells and knocked down Ku86 by specific Ku86 short hairpin RNA (shRNA) in normal and Rad52 À / À B cells, to assess the reciprocal contribution of Rad52 and Ku proteins to CSR. Our findings show that Rad52 competes with Ku70/Ku86 for binding to S-region DSB-free ends to modulate CSR by facilitating a microhomology-mediated S-region DSB A-NHEJ synaptic process. This favours intra-S region recombination, but also mediates Ku-independent inter-S-S region DSB recombination.

Results
Rad52 À / À B cells increase CSR. To test the hypothesis that Rad52, a DNA single-strand annealing (SSA) and HR factor, plays a role in CSR, we analysed B cells from Rad52 À / À C57BL/6 mice and Rad52 þ / þ C57BL/6 littermates for their ability to undergo CSR. Previous report on targeted inactivation of Rad52 showed that Rad52 knockout reduced HR and stated that 'Rad52 is not necessary for CSR 39 . Data, however, were not presented to support the latter claim, which was putatively based on Sm-Se recombination only. We stimulated Rad52 À / À and Rad52 þ / þ B cells, as well as Aicda À / À and Aicda þ / þ B cells with lipopolysaccharide (LPS; to induce CSR to IgG3), mCD154 or LPS plus interleukin (IL)-4 (IgG1), LPS plus interferon (IFN)-g (IgG2c) and LPS plus transforming growth factor (TGF)-b, IL-5, IL-4 and anti-d monoclonal antibody/dex (IgA). After 96 h of culture, the proportion of surface IgG1 þ , IgG3 þ , IgG2c þ and IgA þ B cells among Rad52 À / À B cells was B1.9-, 1.6-, 2.1-and 2.0-folds of those among Rad52 þ / þ B cells (Fig. 1a). This reflected an effective increase of CSR in Rad52 À / À B cells, as these completed the same number of divisions as their Rad52 þ / þ counterparts in response to different doses of LPS or mCD154 plus IL-4, while yielding a 450% increase in switched IgG1 þ B cells (Fig. 1b,c). Increased Rad52 À / À B cell CSR was confirmed by detection of recombinant Sm-Sg1 or Sm-Sg3 DNA (semi-quantitative digestion-circularization PCR; Fig. 1d). Increased Rad52 À / À B-cell CSR was not associated with changes in germline intervening-constant heavy chain region (I H -C H ) transcripts or AID, Ku70/Ku86 and Poly levels, as shown by unchanged Ig1-Cg1, Ig3-Cg3, Aicda, Ku70/Ku86 and Poly transcripts (real-time quantitative reverse transcriptase-PCR (qRT-PCR)) in Rad52 À / À B cells, after a 60 h culture with LPS plus IL-4 or LPS alone (Fig. 1e), as well as unchanged AID, Ku70/Ku86 and Poly protein levels (immunoblotting) after a 72 h culture with LPS plus IL-4 (Fig. 1f). This contrasted with the increased circle Ig1-Cm and Ig3-Cm transcripts, and post-recombination Im-Cg1 and Im-Cg3 transcripts, which are generated only after completion of CSR. Thus, intrinsic B-cell Rad52 deficiency increases CSR in the presence of normal levels of germline I H -C H transcripts and Aicda, Ku70/Ku86 and Poly expression, as well as normal cell division rates.
The error-prone translesion DNA Poly, which, as we have shown, plays a role in SHM 37 , can anneal and end-join DNA single-strand ends, leading to the introduction of microhomologies [34][35][36] , as in the CSR S-S region synaptic process 40 . To determine whether Poly is actually involved in CSR, we stimulated Poly À / À and Poly þ / þ B cells with LPS plus IL-4, LPS only, LPS plus IFN-g and LPS plus TGF-b, IL-5, IL-4 and anti-d monoclonal antibody/dex, to induce CSR to IgG1, IgG3, IgG2c and IgA, respectively. After 96 h of culture, the proportions of class-switched IgG1 þ , IgG3 þ , IgG2c þ and IgA þ Poly À / À B cells were comparable to those of Poly þ / þ B cells, as were B-cell divisions and the proportion of class-switched cells/round of cell division (Fig. 3a,b). The normal CSR in Poly À / À B cells was supported by the normal levels of Aicda expression, circle Ig1-Cm and Ig3-Cm transcripts, post-recombination Im-Cg1 and Im-Cg3 transcripts, as well as germline Ig1-Cg1 and Ig3-Cg3 transcripts (that is, comparable to Poly þ / þ B cells) (Fig. 3c). In addition, Poly À / À B cells were comparable to Poly þ / þ B cells in Sm-Sg1 junction frequency of microhomologies and length (P ¼ 0.09, paired Student's t-test) ( Supplementary Fig. 3). Finally, that Poly À / À B cells functioned normally in CSR was confirmed by analyses in vivo showing comparable levels of serum IgM, IgG1 and IgA, as well as IgG1 þ B220 þ PNA hi GC B cells in the spleens from Poly À / À and Poly þ / þ littermates that had been injected with NP 16 -CGG (Fig. 3d,e). Thus, the DNA single-strand annealing Poly does not apparently play a role in CSR synapsing of S-S region DSBs.

Rad52 deficiency increases class-switched antibody responses.
To define the impact of Rad52 deficiency on CSR in vivo, we analysed the class-switched antibody response to  Supplementary Fig. 1), and Sm-Sa junction sequences with indicated numbers of nucleotide overlaps (microhomologies) in Rad52 þ / þ (n ¼ 30) and Rad52 À / À B cells (n ¼ 30) from the Peyer's patches of three pairs of Rad52 þ / þ and Rad52 À / À C57BL/6 littermates (as in Supplementary  Fig. 2). 0 indicates no microhomology. The average length of nucleotide overlap and the numbers of sequences analysed (n) are indicated. P-values determined using a paired Student's t-test. 4-Hydroxy-3-nitrophenyl acetyl hapten (NP) in Rad52 À / À mice and Rad52 þ / þ littermates after injection of NP-conjugated chicken gamma globulin (NP 16 -CGG), which preferentially induces T-dependent NP-specific IgG1 antibodies. Rad52 À / À mice showed significantly higher titres of NP 32 -binding IgG1 and (high affinity) NP 4 -binding IgG1, as well as total IgG1 and IgA than their Rad52 þ / þ counterparts, in the presence of normal IgM levels (Fig. 4a). In addition, similar Rad52 À / À mice showed significantly higher titres of total IgE than their Rad52 þ / þ littermates after injection of ovalbumin (OVA; Fig. 4b). In Rad52 À / À mice, high IgG1 and IgA, and NP-binding IgG1 titres were associated with a 475% increase of IgG1 þ B220 þ PNA hi GC B cells in the spleen (Fig. 4c) and a 100% increase in IgA þ B220 þ PNA hi GC B cells in Peyer's patches. In Rad52 À / À mice, the spleen size, and number and size of Peyer's patches were comparable to those in their Rad52 þ / À and Rad52 þ / þ counterparts. Consistent with a previous report 39 , the number of B (B220 þ ) and T (CD3 þ ) cells, and the proportions of CD4 þ and CD8 þ T cells in Rad52 À / À mice were comparable to those in their Rad52 þ / þ littermates. In addition, the proportion of B220 þ PNA hi GC B cells and B220 low CD138 þ plasma cells in the spleen were also comparable in Rad52 À / À and Rad52 þ / þ mice; Rad52 À / À mice showed normal B-cell viability (Fig. 4d), proliferation and normal cell cycle in total and GC B cells, as measured by bromodeoxyuridine (BrdU) incorporation and 7-aminoactinomycin D (7-AAD) staining (Fig. 4e). Thus, Rad52 deficiency leads to significantly increased class-switched antibody response without affecting B-cell proliferation or survival.
Rad52 competes with Ku70/Ku86 for binding to S-region DSB ends. The increased CSR in Rad52 À / À B cells in vitro and in vivo suggested that Rad52 is recruited to S-region DSB-ends where it would compete with Ku70/Ku86 to modulate CSR. To demonstrate that Rad52 and Ku70/Ku86 indeed bind to CSR-targeted S regions, we performed ChIP assays with anti-Rad52 antibody and anti-Ku70/Ku86 monoclonal antibodies. These showed that similar to Ku70/Ku86, Rad52 was specifically recruited to Sm and Sg1 regions, not Sg3, in B cells stimulated by LPS plus IL-4 (undergoing CSR to IgG1), and to Sm and Sg3, not Sg1, in B cells stimulated by LPS (CSR to IgG3). Rad52 and Ku70/Ku86 could be readily detected on Sm, and Sg1 or Sg3, but not Cm region, in Aicda þ / þ B cells stimulated with LPS plus IL-4 or LPS alone, but failed to associate with such S regions in similarly activated Aicda À / À B cells, showing that similar to Ku70/Ku86, recruitment of Rad52 to CSR-targeted S regions was dependent on S region AID processing (Fig. 5a,b). To analyse the impact of Rad52 on Ku70/Ku86 recruitment to CSR-targeted S regions, we performed ChIP assays using anti-Ku70/Ku86 monoclonal antibody in Rad52 þ / þ and Rad52 À / À B cells stimulated with LPS plus IL-4 or LPS. We complemented this approach with in situ DNA end-labelling by biotin-16-dUTP (bio-dUTP) 41 followed by ChIP with anti-Ku70/86 monoclonal antibody or anti-Rad52 antibody and capture of biotin-labelled DNA fragments with streptavidin magnetic beads. This approach, which allows for detection of Ku70/Ku86 or Rad52 on DSB ends, showed that Rad52 indeed bound to the free ends of DSBs in the CSR-targeted S regions, but not to Cm region or S regions not involved in CSR. In the absence of Rad52, Ku70/Ku86 bound to CSR-targeted Sg1 and Sg3 DSB free ends, up to 7.6-fold more than in the presence of Rad52 (Fig. 5c,d).
Rad52 À / À B cells reduce intra-S region DNA recombination. AID generates multiple DSBs within the targeted S regions, many of which are rejoined or joined to other DSBs within the same S region 12 . Each S region consists of highly repetitive DNA motifs. These can give rise to complementary protruding ends in upstream and downstream DSBs that are suitable substrates for DNA annealing by Rad52. Subsequent synapse of such DSB ends leads to intra-S region recombination and deletion of intervening DNA (of variable length). All S regions include characteristic and highly repetitive motifs. Such highly repetitive motifs differ in both nature and frequency in different S regions. It follows that DSB protruding ends from two different S regions, such as Sm and Sg1, will encompass sequences of virtually no complementarity, making them poor substrates for Rad52-mediated complementary annealing than protruding ends from DSBs within the same S region. Indeed, Pustell Matrix dot plot analysis of human and mouse Sm and Sg1 revealed maximal sequence complementarity within the individual Sm or Sg1 region, in particular within their core region, and maximal lack of complementarity between these two S regions, in particular in their core sequences (Fig. 7a). Accordingly, the reduced CSR in Ku70 deficiency, which, as we have hypothesized, would result in increased Rad52 recruitment to S-region DSB-ends, was shown to be associated with significantly increased occurrence of intra-S region deletions, remnants of occurred intra-S region recombinations 17 .
We hypothesize here that such increased intra-S region deletions are mediated by a DSB synaptic process involving Rad52. If our hypothesis is correct, then the absence of Rad52which, as shown by the preceding experiments, led to increased recruitment of Ku70/Ku86 to S-region DSB ends and increased CSR-would result in reduction of intra-S region deletions. To test our hypothesis, we set up to assay for deletions within the Sm region of genomic DNA from Rad52 þ / þ and Rad52 À / À B cells activated for CSR to IgG1-intra-S region deletions occur far more frequently in Sm than the other (downstream) S regions 17using specific primers flanking both sides of Sm to amplify this DNA region. Sm DNAs with internal deletions were detected by visualizing amplification products that were shorter than germline Sm and then positively identified by DNA sequencing (Fig. 7b,c and Supplementary Fig. 4). This revealed that Rad52 À / À B cells displayed a substantially lower frequency of intra-Sm region deletions as compared with their Rad52 þ / þ counterparts (4 out of 30, that is, 13.3%, versus 14 out 30, that is, 46.7%, P ¼ 0.0006, paired Student's t-test). Even taking into account a remote possibility that some of the S-region DNA deletions might have been stemmed from PCR amplification or propagation of S region-containing plasmids in bacteria, these findings indicate that the Rad52-mediated DSB synaptic process favours intra-S region over inter-S-S region recombination.
Rad52 and Ku86 reciprocally modulate CSR. To define whether, in addition to mediating intra-S region recombination, Rad52 can also mediate inter-S-S region recombination, we knocked down Ku86 in Rad52 À / À and Rad52 þ / þ B cells using a pGFP-C-Ku86-shLenti lentiviral vector. Expression of a Ku86specific shRNA reduced Ku86 protein level by 87% (average) (when taking into consideration the 80% lentiviral transduction efficiency in our experiments) without altering Rad52 protein expression (Fig. 8a)-a pGFP-C-scrambled-shLenti lentiviral vector, which did not alter Ku86 expression, was used as a control. The transduced B cells were cultured with LPS and IL-4 for 96 h before analysing CSR to IgG1. Similar to untransduced Rad52 À / À B cells, pGFP-C-scrambled-shLenti lentiviral vectortransduced Rad52 À / À B cells displayed a (54%) higher level of CSR to IgG1 (P ¼ 0.003, paired Student's t-test) as compared with their Rad52 þ / þ counterparts (Fig. 1a,b). In similar Rad52 À / À B cells, Ku86 knockdown by pGFP-C-Ku86-shLenti lentiviral vector resulted in a nearly tenfold reduction of CSR to IgG1, as compared with the twofold reduction in Rad52 þ / þ B cells transduced by the pGFP-C-Ku86-shLenti lentiviral vectorknockdown of Ku86 in Rad52 þ / þ B cells reduced CSR to an extent comparable to that reported for Ku70-or Ku86-deficient B cells 18 . The less profound CSR reduction by Ku86 knockdown in Rad52 þ / þ than Rad52 À / À B cells reflected the contribution of Rad52 to inter-S-S region recombination in Rad52 þ / þ B cells-the residual 3.4% CSR in pGFP-C-Ku86-shLenti lentiviral vector-transduced Rad52 À / À B cells was (probably) due to the residual (B12%) functional Ku86 in these B cells (Fig. 8a). Although knockdown of Ku86 in Rad52 À / À B cells virtually ablated CSR, knockdown of Ku86 in Rad52 þ / þ B cells resulted only in partial CSR reduction, thereby pointing at a contribution of Rad52 to inter-S-S region synapses in the absence of Ku86. Thus, Rad52 partially rescues CSR in B cells with a compromised Ku-dependent/NHEJ repair pathway.
Our experiments have shown that in the absence of Ku86, Rad52 can mediate inter-S-S region synapses. In the presence of Ku70/Ku86, Rad52 would compete with this heterodimer for binding to S-region DSB ends, thereby skewing the S-region DSB synaptic process towards intra-S region recombination to the detriment of inter-S-S region recombination and CSR. To test the hypothesis that overexpression of Rad52 will further reduce recruitment of Ku70/Ku86 to S-region DSB ends and significantly dampen CSR, we transduced Rad52 þ / þ Ku86 þ / þ (normal) B cells with pMIG-Rad52 or an empty pMIG retroviral vector as control, cultured them with LPS plus IL-4 for 96 h, before analysing CSR. Enforced Rad52 expression in Rad52 þ / þ Ku86 þ / þ B cells increased Rad52 protein level by 2.4-fold without altering Ku86 protein expression and reduced by 465% CSR to IgG1 (Fig. 8a,c). Thus, Rad52 (in excess) competed with Ku70/Ku86 for binding to S-region DSB ends, thereby skewing the DSB synaptic process towards intra-S region recombination and significantly decreasing CSR.
Switching B cells decrease Rad52 and increase Ku70/Ku86. Given the reciprocal modulation of CSR by Rad52 and Ku70/ Ku86, we hypothesized that, on exposure to CSR-inducing stimuli, B cells downregulate Rad52 and upregulate Ku70/Ku86 to reduce intra-S region and facilitate inter-S-S region recombination, thereby ensuring maximal CSR rates. Indeed, in B cells stimulated with increasing amounts of LPS and LPS or mCD154 plus IL-4, which induced Aicda and CSR, Rad52 transcript levels were reduced by 73%-96% (Po0.0001, paired Student's t-test) after 24-48 h of culture (Fig. 9a). In contrast to Rad52, Ku70 and Ku86 transcripts were increased by two-to threefolds (P ¼ 0.00013 or Po0.0001, respectively, paired Student's t-test) within 24 h and returned to baseline values by 48 h. Poly followed the upregulation/downregulation kinetics of Ku70/Ku86. B cells from p53 À / À Rad52 þ / þ and p53 À / À Rad52 À / À mice were stimulated with LPS plus IL-4 for 96 h before genomic DNA isolation. (a) c-Myc/IgH translocations were identified by amplifying DNA using long-range nested PCR involving primers specific to the IgH and c-Myc locus, and verified by Southern blot hybridization with an IgH or c-Myc-specific probe. Each PCR assay was performed using template DNA from 10 6 cells. Twenty-five amplicons from p53 À / À Rad52 þ / þ B cells and 25 amplicons from p53 À / À Rad52 À / À B cells are shown. The frequencies of c-Myc/IgH translocations per cell are indicated below the gel images. PCR amplification products that can be detected by both IgH and c-Myc DNA probes were from c-Myc/IgH translocations: 8 of 25 and 3 of 25 amplicons from p53 À / À Rad52 þ / þ and p53 À / À Rad52 À / À B cells, respectively, contained PCR amplification products from c-Myc/IgH translocations.  (Fig. 9b). Thus, B cells induced to undergo CSR modulate Rad52 (RAD52) and Ku70/Ku86 (KU70/KU86) expression in a reciprocal manner at both transcription and protein levels, thereby skewing the S-region DSB synaptic process towards inter-S-S region recombination.

Sμ Sμ
Figure 7 | Rad52-mediated S-region DSB repair favours intra-S region DNA recombination. (a) Each S region consists of highly repetitive motifs, which can facilitate the formation of microhomologies, in particular within the S region core. As the characteristically repetitive sequences are virtually unique to S regions, DSB ends in the same S region are better suited substrates for Rad52-mediated complementary DNA single-strand annealing than those in two different S regions, such as Sm and Sg1. Repetitive sequence elements in human and mouse Sm and Sg1 that can potentially form microhomologies were identified by Pustell Matrix dot plot using MacVector software and are depicted by small dots. Thick lines indicate the core regions of Sm and Sg1. (b) Schematic representation of the detection of intra-S region recombination (deletion) in Sm region by PCR amplification. DNA-amplified sequences of Sm region that underwent intra-S region DNA recombination are shorter than those of Sm in the germline configuration. (c) Rad52 þ / þ and Rad52 À / À B cells were stimulated with LPS plus IL-4 for 96 h. Sm region DNA was amplified by nested PCR. PCR amplification products were then cloned into the TOPO cloning vector. Sm region sequences from individual clones amplified by PCR and resolved through a 0.8% agarose gel. PCR amplification products smaller than that amplified from the germline Sm region DNAs (indicated by arrows) are from Sm region DNAs that underwent intra-S region recombination, thereby deleting variable lengths of DNA: 14 of 30 Sm region DNAs in Rad52 þ / þ B cells and 4 out of 30 Sm region DNAs in Rad52 À / À B cells underwent intra-S region recombination. Data are from three pairs of Rad52 þ / þ and Rad52 À / À C57BL/6 mice. generally repaired by HR or NHEJ-NHEJ is also referred to as classical-NHEJ. HR accurately repairs staggered post-replicative DSBs using a long, homologous single-strand template in the form of a sister chromatid, whereas NHEJ fuses blunt or virtually blunt DSB ends that lack substantial joining complementarity, to form 'direct' junctions 10,12 . Rad52 and Ku70/Ku86 are critical and early elements in HR and NHEJ DSB, respectively. Both Rad52 and Ku70/Ku86 bind to DSB ends and facilitate end-to-end interactions 43 . Rad52 binds to and wraps around resected DSBs, thereby efficiently promoting annealing of complementary DNA single strands 31 , although it can also bind to blunt DSB ends [43][44][45] . Ku70/Ku86 binds to blunt DSB ends and gives rise to 'clean' (that is, free of microhomologies) DSB junctions, although it can also bind, albeit less efficiently, to DSBs with relatively short protruding ends 44,46 .
Efficient inter-S-S region recombination that leads to CSR requires DSB resolution by NHEJ, which recombines S-region DSBs to form S-S junctions with no or minimal microhomologies [10][11][12] . Substantial CSR, however, occurs in the absence of core NHEJ components by A-NHEJ, which yields S-S junctions with microhomologies 12,[16][17][18] . In addition to CSR, microhomology-mediated A-NHEJ would also effect oncogenic inter-chromosomal translocations, which are indeed characterized by DSB junctions with significant levels of microhomologies 8,25 . Here we show that the SSA and HR element Rad52 plays a central role in facilitating the A-NHEJ DSB synaptic process underpinning CSR, as well as t(15;12) c-Myc/IgH translocation. Rad52 functions as a recombination mediator and is involved in maintenance of genomic integrity. It interacts with replication protein A 47 , another single-strand DNA-binding protein, which plays a role in CSR. Replication protein A associates preferentially with resected single-strand DNA ends created by AID, thereby helping stabilize AID itself 48 .
Extensive resection of DSBs yielding long overhangs (200 nt to 2-4 kb) 8,9 leads to HR, whereas less extensive DSB resection (generally 430 but o200 nt) 20,32 promotes what is referred to as SSA pathway, which is mediated by Rad52 but does not require a sister chromatid template 49,50 . More limited (o30 nt) DSB resection facilitates single-strand DNA annealing leading to MMEJ 9,20,48 , suggesting that the distinction between SSA and MMEJ is based on the length of the complementary protruding ends involved in the synaptic process. The requirement for Rad52 in SSA but not MMEJ has been claimed to be another criterion to distinguish SSA from MMEJ 21 . Data in fission yeast, however, suggest an extensive overlap between SSA and MMEJ, as effected by Rad22, the fission yeast homologue of Rad52 (refs 23,27). Other reports have attributed contradictory roles to Rad52 in MMEJ. Rad52 has been shown to suppress MMEJ in budding yeast, possibly by promoting Rad51 filament formation and thereby HR 51,52 , but it has also been shown to be required for MMEJ in both budding and fission yeast 22,23,53 . DSB junctional microhomologies in yku70/80Drad52D budding yeast cells were thought to indicate an independence of MMEJ from Rad52. Such microhomologies, however, were likely to be introduced by Rad59, a Rad52 homologue that anneals single-strand DNAs with short homologous sequences 23,27,45,54 .
By virtue of its low-fidelity translesion DNA synthesis, Poly plays an important role, as we have shown, in SHM 37 . Poly À / À B cells were comparable to their Poly þ / þ counterparts in switching to IgG1, IgG3, IgG2c and IgA, as well as in frequency and extent of Sm-Sg1 junctional microhomologies. These findings confirm and extend a previous observation 40 , and further suggest that Poly does not play a role in Ku-independent (microhomology-mediated) A-NHEJ of CSR. This is in spite of the ability of Poly to promote annealing of resected DSBs and facilitate MMEJ 32,34-36 . The failure of Poly deficiency to alter Redundancy in A-NHEJ would not be limited to Poly, as deficiency of Lig1 or Lig3 (the DNA ligases involved in A-NHEJ) alone did not affect CSR, possibly due to the functional interchangeability of these two ligases in the late stage of CSR 56 . Unlike Poly, Rad52 plays an important role in CSR. Similar to Ku70/Ku86, Rad52 was recruited to CSR-targeted S regions after AID processing. Increased CSR in the absence of Rad52 entailed enhanced recruitment of Ku70/Ku86 to S-region DSB ends, decreased intra-Sm DSB recombinations, reduced frequency and length of microhomologies in S-S junctions, as well as decreased inter-chromosomal c-Myc/IgH translocations. All of these features are opposite to those in (NHEJ-deficient) Ku70 À / À B cells, which showed increased intra-S region DSB recombination, increased frequency and length of microhomologies in S-S junctions, as well as increased interchromosomal c-Myc/IgH translocations [16][17][18] , and point at Rad52 as mediator of A-NHEJ in CSR (Supplementary  Table 1) 17,18 . Thus, our findings argue for a role of Rad52 in a Ku-independent DSB synaptic process that gives rise to S-S junctions. These contain microhomologies, which are more abundant than those, if any, of Ku-dependent S-S synapses 16-18 . Recruitment of Rad52 or Ku70/Ku86 to DSB ends would lead to different synaptic processes 43 . Rad52 preferentially synapsed DSB ends within the same S region, as emphasized by the greatly decreased intra-Sm region deletions, which are marks of intra-S region DNA recombination or 'rejoining', in Rad52 À / À B cells. Conversely, intra-S region deletions were found to be greatly increased in B cells lacking Ku70 or Ku70 and Lig4 (ref. 17 protein that favours long-range S-S recombination (leading to CSR), but prevents short-range rejoining of intra-S region DSBs, possibly because of its protection of DNA ends from resection 57 . Rad52 recruitment to CSR-targeted Sg was as efficient as recruitment to Sm region. However, because of the distinctive, albeit characteristically repetitive, sequences of the IgH locus S regions, DSB protruding ends in two different S regions, such as Sm and Sg1, are less suitable substrates for Rad52-mediated annealing than DSB protruding ends in the same S region. Thus, although (Ku-dependent) NHEJ skewed the DSB repair process towards inter-S-S region recombination, thereby increasing CSR rates, Rad52-dependent annealing of DNA single strands favoured DSB synapsing within the same S region, which led to intra-S region recombinations, thereby lowering CSR rates. The latter process was further emphasized by the dampening of CSR on enforced expression of Rad52. We show here that, similar to Ku70/Ku86, Rad52 bound to the DSB free ends of CSR-targeted S regions, but did not bind to S regions not involved in CSR or Cm region. By binding to S-region DSB free ends, Rad52 prevented Ku70/Ku86 from accessing such DSB ends, as shown by the specific ChIP assays involving anti-Ku70/Ku86 monoclonal antibody and Rad52 À / À B cells, the specific competition assays using recombinant Rad52 and Ku70/Ku86 proteins, as well as the supershift EMSAs involving anti-Ku70/Ku86 monoclonal antibody and nuclear extracts from Rad52 À / À B cells. The competition of Rad52 with Ku70/Ku86 for binding to S-region DSB ends was further emphasized by the decreased CSR on enforced expression of Rad52 in (normal) B cells. Such CSR decrease was more pronounced than that resulting from knocking down Ku86 in normal B cells. In these B cells, Ku86 was present in residual amounts but competed with lower amounts of Rad52 than in B cells in which Rad52 was enforced expressed. The virtual ablation of (the otherwise much increased) CSR by Ku86 knockdown in Rad52 À / À B cells not only further emphasized the critical role of the Ku70/Ku86 heterodimer in the inter-S-S region DSB synaptic process but also revealed a contribution of Rad52 to inter-S-S region recombination and CSR.
In B cells induced to undergo CSR, execution of the DSB synaptic process towards inter-S-S region recombination would be further facilitated by the downregulation of Rad52 (RAD52) with concomitant upregulation of Ku70/Ku86 (KU70/KU86) together with Aicda (AICDA) and Poly (POLy) transcripts at early time points (24-48 h). The expression of Rad52, Ku70/Ku86, AID and Poly proteins followed tightly the kinetic expression of their corresponding transcripts. This was particularly the case for Rad52, which is a short-lived protein 58 , in contrast to the relatively stable Ku70/Ku86. In switching B cells, Rad52 (RAD52)/Rad52 downregulation was profound but not complete, possibly allowing for some residual intra-S recombination. After 48-72 h, the return of Ku70/Ku86 (KU70/ KU86) transcripts and Ku70/Ku86 proteins to a lower expression levels would be important to maintain a Rad52:Ku70/Ku86 ratio that prevents CSR dysregulation. Thus, the modulatory changes in Rad52 and Ku70/Ku86 are likely to be the result of CSR induction rather than a consequence of cell proliferation.
As we showed here, Rad52 deficiency did not alter B cell cycle or B-cell division and increased CSR independently of cell division. Because of its use of sister chromatids as templates to mediate faithful repair, HR is generally restricted to S and G2 phases. NHEJ and MMEJ generally operate in G0-G1 8 , although both can function throughout the whole cell cycle 8 . In yeast, Rad52 is generally recruited to DSBs and readily form foci in response to DNA damage in S and G2/M, when a sister chromatid template becomes available to guide the repair process 59 . Rad52, however, can also accumulate at DNA damage sites immediately after insertion of DSBs outside the S and G2/M phases 60,61 , thereby promoting annealing of singlestrand DNA and leading to (micro)homology-mediated endjoining in the absence of a sister chromatid template. Furthermore, HR has been suggested to contribute to S-region DSB repair and CSR, in addition to NHEJ/A-NHEJ 48,62 . Indeed, DSBs not processed by NHEJ or A-NHEJ in G0-G1 would be further resected in S-G2/M and repaired by HR using the second (intact) IgH allele as a template 48 . Thus, given the role of Rad52 in both A-NHEJ and HR, the contribution of Rad52 to S region DSB repair would extend throughout the whole cell cycle. Rad52facilitated microhomology-mediated A-NHEJ may complement NHEJ in the presence of high DSB rates, such as those in hypermutating/switching GC B cells, thereby functioning as a salvage pathway for DSBs that are not repaired by NHEJ.
Collectively, our findings show that Rad52 competes with Ku for binding to S-region DSB free ends, where it facilitates a DSB synaptic process, which favours intra-S region recombination, and also mediates, in particular in the absence of a functional NHEJ pathway, inter-S-S region recombinations ( Supplementary  Fig. 5).

Methods
Mice. Rad52 À / À mice were generated by Dr Albert Pastink (Leiden University, Leiden, The Netherlands) by replacing exon 3 of the Rad52 gene with positive selection marker neomycin, as driven by the phosphoglycerate kinase promoter, and an upstream mouse sequence functioning as a transcription terminator 39 . These mice were backcrossed to C57/BL6 mice for more than six generations. No full-length or truncated Rad52 protein was produced from the disrupted allele. Rad52 À / À mice were viable and fertile, and showed no gross abnormalities 39 . p53 À / À (B6.129S2-Trp53 tm1Tyj /J) mice were purchased from Jackson Laboratory (Bar Harbor, Maine). p53 À / À Rad52 þ / þ and p53 À / À Rad52 -/mice were generated in our facilities by cross-breeding p53 À / À with Rad52 þ / À mice. Aicda À / À mice (C57BL/6 background) 63 were obtained from Dr Tasuku Honjo (Kyoto University, Kyoto, Japan). Poly À / À mice (C57BL/6 background) were generated by Dr John Schimenti (Cornell University, Ithaca, NY) by placing an in-frame stop codon into exon 1 and replacing exons 2-5 with a neomycin resistance (neo) gene 37 . All mice were housed in pathogen-free conditions. Both male and female mice aged 8-12 weeks were used for the experiments. The Institutional Animal Care and Use Committee of the University of Texas Health Science Center at San Antonio approved all animal protocols. NP 16 -CGG and OVA immunization and antibody titration. Rad52 þ / þ , Rad52 À / À , Poly þ / þ and Poly À / À mice (8-12 weeks of age) were injected intraperitoneally (i.p.) with 100 mg of NP 16 -CGG (average 16 molecules of 4-hydroxy-3-nitrophenyl acetyl coupled to 1 molecule of chicken g-globulin; Biosearch Technologies) in 100 ml of alum (Imject Alum, Thermo Fisher Scientific). Serum was collected 10 days later for titration of circulating total and NP-binding IgM and IgG1 using enzyme-linked immunosorbent assays, as we described [64][65][66] . To analyse the impact of Rad52 deficiency in IgE response, Rad52 þ / þ and Rad52 À / À mice were injected i.p. with 20 mg of OVA in 100 ml of alum and 'boost' injected 7 days later with 20 mg of OVA in PBS. Serum was collected before or 5 days after the 'boost' injection for titration of circulating total IgE. B and T cells, cell cycle and proliferation. B cells (B220 þ ), CD4 þ and CD8 þ T cells, dead B cells, GC (B220 þ PNA hi ) B cells and plasma cells (B220 lo CD138 þ ) were identified and analysed by flow cytometry 64,65,67 using a FACSCalibur or LSR-II flow cytometer (BD Biosciences). Data analysis was performed using FlowJo software (Tree Star). Single-cell suspensions were prepared from spleens of 30 min at room temperature with anti-Ku70/Ku86 monoclonal antibody (MA1-21818, Thermo Fisher Scientific, 1 mg per reaction) or irrelevant mouse IgG before being incubated with the biotin-labelled Sm probe. Electrophoresis of the DNA-protein complexes was carried out on a 5% non-denaturing polyacrylamide gel with 0.5 Â TBE. The samples were then transferred to Hybond-N þ membranes with a semi-dry transfer followed by ultraviolet cross-linking. Detection was performed using the Chemiluminescent Nucleic Acid Detection Module (Thermo Fisher Scientific) according to the manufacturer's instructions.
c-Myc/IgH translocations. Genomic DNA was isolated from p53 À / À Rad52 -/and p53 À / À Rad52 þ / þ B cells stimulated with LPS plus IL-4 for 96 h. Nested PCRs for translocations were performed on genomic DNA from 1 Â 10 6 cells with GoTaq Long PCR Master Mix (Promega) using IgH forward and c-Myc reverse primers (Supplementary Table 2). PCR conditions were as follows: 95°C for 2 min followed by 25 cycles (95°C, 10 s; 62°C, 45 s; and 65°C, 6 min) for both the first and second rounds. Amplified DNA were fractionated through 1.0% agarose, blotted onto Hybond-N þ membranes and hybridized to biotin labelled c-Myc-specific oligonucleotide probe 5 0 -GACGCCACTGCACCAGAGACCCTGC AGCGATTCAG-3 0 or IgH-specific oligonucleotide probe 5 0 -CCTGGTATACAG GACGAAACTGCAGCAG-3 0 and detected using Streptavidin-HRP. To analyse the c-Myc-IgH junctional sequences, PCR products were treated with T4 DNA polymerase in the presence of dNTP, cloned into the pCR-Blunt II-TOPO vector (Invitrogen) and sequenced. Sequence alignment was performed by comparing the sequences of PCR products with germline c-Myc and IgH genomic sequences using National Center for Biotechnology Information BLAST.
Analysis of intra-Sl region recombination. To analyse intra-Sm region DNA recombination, Sm region DNA was amplified by nested PCR using mouse Sm-specific primers (Supplementary Table 2) and GoTaq Long PCR Master Mix (Promega) from genomic DNA isolated from Rad52 þ / þ and Rad52 À / À B cells stimulated with LPS plus IL-4 for 96 h. Amplified DNA was treated with T4 DNA polymerase in the presence of dNTP and cloned into the pCR-Blunt II-TOPO vector (Invitrogen). Sm region DNA of individual clones were re-amplified using the Sm-specific primers (as above) and GoTaq Long PCR Master Mix. Amplified DNA was fractionated through 0.8% agarose gel. The intra-Sm recombined DNAs, that is, Sm region DNA with internal deletions, were detected by comparing the length of the amplified DNA with that of the respective germline Sm region and positively identified by sequencing.
Rad52 retroviral construct and enforced expression. Rad52 cDNA was amplified from mouse B cells using the appropriate primers (Supplementary Table 2) and cloned into the pMIG retroviral expression vector (green fluorescent protein (GFP) translation initiated by the internal ribosome entry sites (IRES) in transduced B cells). To generate the retrovirus, pMIG vector encoding GFP or pMIG-Rad52 construct encoding GFP and Rad52, together with the pCL-Eco retrovirus-packaging vector (Imgenex), were used to transfect HEK293T cells by a Ca þ þ phosphate-mediated transfection procedure (ProFection Mammalian Transfection System, Promega). Viral supernatants were collected and used to transduce spleen B cells from C57BL/6 mice, as we reported 64,65 , after a 12 h LPS activation. Transduced B cells were then stimulated with LPS plus IL-4 for 96 h before analysis of GFP þ and IgG1 þ B cells by flow cytometry 64,65 -dead (7-AAD þ ) cells were excluded from analysis. Expression of Rad52, Ku86 and b-Actin proteins in B cells transduced with empty or Rad52-expression pMIG retroviral construct were analysed by immunoblotting 66 .
Statistical analysis. Statistical analysis was performed using Excel (Microsoft), to determine P-values by paired and unpaired Student's t-tests; and P-values o0.05 were considered significant. Differences in the number and length of microhomologies in Sm-Sg1 and Sm-Sa junctions between stimulated Rad52 þ / þ and Rad52 À / À , or Poly þ / þ and Poly À / À B cells were analysed using the paired Student's t-test.
Data availability. The data that support the findings of this study are available within the article and its Supplementary Information file or from the corresponding authors upon reasonable request.