Efficient AID targeting of switch regions is not sufficient for optimal class switch recombination

Antibody affinity maturation relies on activation-induced cytidine deaminase (AID)-dependent somatic hypermutation (SHM) of immunoglobulin (Ig) loci. Class switch recombination (CSR) can in parallel occur between AID-targeted, transcribed, spliced and repetitive switch (S) regions. AID thus initiates not only mutations but also double-strand breaks (DSBs). What governs the choice between those two outcomes remains uncertain. Here we explore whether insertion of transcribed intronic S regions in a locus (Igκ) strongly recruiting AID is sufficient for efficient CSR. Although strongly targeted by AID and carrying internal deletions, the knocked-in S regions only undergo rare CSR-like events. This model confirms S regions as exquisite SHM targets, extending AID activity far from transcription initiation sites, and shows that such spliced and repetitive AID targets are not sufficient by themselves for CSR. Beyond transcription and AID recruitment, additional IgH elements are thus needed for CSR, restricting this hazardous gene remodelling to IgH loci. Activation-induced cytidine deaminase can induce both somatic hypermutation (SHM) and class switch recombination in immunoglobulin loci. Here the authors show that spliced and repetitive switch regions are exquisite SHM targets, but are not sufficient by themselves for class switch recombination.

C lass switch recombination (CSR) occurs at immunoglobulin heavy chain (IgH) loci after mature B-cell activation 1 . By changing the IgH chain constant (C H ) gene, it switches from IgM/IgD production to either IgG, IgA or IgE 2 . At the gene level, CSR needs 'switch' (S) regions located 5 0 of C H genes to undergo double-strand breaks (DSBs), before donor (Sm) and acceptor (Sg, Sa or Se) regions can be joined and bring a downstream C H gene in place of Cm. CSR and occurrence of DSBs require adequate chromatin structure, looping of IgH loci 3 and transcription of target C H genes within specific units associating a I-promoter, a non-translatable 'I-exon', an intronic S region and the coding C H exons 1 . Although its role is still unclear, this germline transcription (GLT) initiating from I-promoters and running through S and C regions is mandatory for CSR and S region accessibility to the activation-induced cytidine deaminase (AID) enzyme 4 . S regions are 1-10 kb-long highly repetitive DNA sequences containing RGYW motifs targeted by AID 5 . AID, also implicated in the somatic hypermutation (SHM) of V(D)J exons, catalyses cytosine (C) deamination into uracyl (U), such lesions being further processed by either the base excision repair (BER) or mismatch repair (MMR) machinery 5 . G-rich S regions may form R-loops during their transcription, allowing accessibility of the non-template strand to the enzymatic activity of AID 6 . While GLT is non-productive due to stop codons in all three reading frames of I-exons, CSR needs I-exon splicing onto C H exons 7 . This might be due to the contribution of the splicing regulator PTBP2 to AID recruitment 8 . Recent evidence also suggested that, during CSR, AID directly recruits MMR and BER factors 2 . Processing of G:U mismatches by either MMR or BER initiates DSBs, then repaired by classical or to a lesser extent alternate non-homologous end joining (c-NHEJ or a-NHEJ, respectively) 9 .
Besides the architecture of transcriptional units, a longstanding issue about CSR is the role of IgH cis-regulatory elements. If the intronic Em enhancer has only a modest role during CSR 10 , the IgH 3 0 RR lying downstream of the last Ca exon, has long been described as a master regulator of CSR and recently appeared as also mandatory for SHM, its complete deletion in mice abrogating both processes [11][12][13] . The 3 0 RR deletion strongly decreases GLT of some C H genes and modestly affects pV H transcription. The nearly complete V H DJ H SHM blockade in 3 0 RR-deleted mice, suggests that the 3 0 RR also directly helps recruiting AID to V(D)J exons 13 . Whether the 3 0 RR also specifically attracts the CSR machinery onto S regions and/or promotes IgH locus bending and S-S synapses remains to be determined. Finally, although it was proposed that staggered single-strand breaks in repetitive S regions can initiate DSBs 14 , it is unclear whether solely the S region structure controls the choice between processing AID attacks into either single-strand lesions and SHM or the occurrence of DSBs followed with long distance deletions.
To determine the minimal requirements for the induction of DSBs and CSR, we created a model whereby transcribed S regions were introduced into the Igk locus normally targeted by SHM. This 'Knock-In Kappa Switch' (KIKS) mice expressed a functional human Ig Ck exon (hCk) flanked by two transcribed S regions, downstream of the last Jk and thus of endogenous rearranged VkJk exons. Hence, a complete CSR substrate was placed at sites naturally targeted by AID during SHM, while its CSR efficiency could be monitored by following the switch from hCk to murine Ck (mCk) expression. Expression of hCk supported B-cell development in KIKS mice, as previously reported in another mouse featuring hCk expression 15 . However, despite efficient transcription and AID targeting of S regions, the KIKS locus only displayed minimal CSR in few B cells within Peyer's patches (PP). This clearly uncouples the two issues of AID attacks, SHM and CSR. It demonstrates the requirement for specific IgH cis-acting elements which, beyond transcription and AID recruitment onto repetitive targets, finally promote efficient CSR at the IgH locus.

KIKS locus description and B-cell development in KIKS mice.
To determine the minimal requirements for the occurrence of DSBs and CSR after AID DNA lesions, we inserted core Sm and Sg3 regions within the Igk locus. These S regions were inserted into the spliced introns of tandem transcriptional units, but with promoter, enhancer and intron sequences from Igk origin. This strategy avoided any possible influence of IgH cis-acting elements other than S regions on the artificial KIKS locus, although its architecture mimicked the IgH locus, with S regions comprised into the spliced introns of tandem transcriptional units ( Supplementary Fig. 1a). Thus, transcription of the KIKS donor Sm was driven by the promoter of the rearranged Vk, while constitutive transcription of the acceptor Sg3 was ensured by a second knocked-in pVk promoter. Sg3 is also preceded by a small non-coding exon (leader Vk4-63 with an inserted stop codon but retaining its natural donor splice site) to ensure its splicing onto the mCk exon and mimic non-productive IgH GLTs (Fig. 1). This strategy also forbids translation of any truncated and potentially harmful mCk-LC. AID was expected to target the KIKS Sm since it stands immediately downstream of rearranged VkJk exons. Regarding the acceptor Sg3, insertion of an additional pVk in the Igk locus was previously shown to recruit   16 . Before any Sm-Sg3 recombination, VJ-rearranged KIKS loci encode a chimeric 'hk-LC' combining a murine V with a human C domain. If CSR-like events occur, Sm-Sg3 recombination should remove the human Ck and yield a fully murine k-LC (mk-LC; Fig. 1). B-cell development was normal in homozygous KIKS mice regarding bone marrow pro-, pre-and immature B cells and proportions of splenic marginal zone and follicular (Fo) B cells (Supplementary Table 1), with percentage of k-expressing B cells identical to wild-type (WT) mice ( Supplementary Fig. 1b, left panel). The efficient association of the hk-LC with murine HC was demonstrated by the similar membrane expression level of IgM and IgD in KIKS compared with WT B cells ( Supplementary Fig. 1b, right panel). KIKS serum k-LC levels were also similar to WT mice (mean±s.e.m.: 3.72 ± 0.44 versus 4.80 ± 0.45 mg ml À 1 , P ¼ 0.15, respectively, Supplementary Fig. 1c). The KIKS Igk locus thus supports normal B-cell development and function in mice, as observed in another model 15 . Our reporter system was designed to easily monitor CSR-like recombination within the KIKS locus by detecting a switch from human to murine k-LC at the B-cell surface. In non-immunized KIKS mice, mk-LC-expressing B cells were below significant detection by flow cytometry ( Supplementary Fig. 1b, right panel). The few cells staining for both human and murine k-LC never exceeded background levels set in parallel with nonspecific isotype controls ( Supplementary  Fig. 1d). Accordingly, we only noticed minimal amounts of mk-LC in sera of KIKS mice (0.017 ± 0.004 mg ml À 1 ; Supplementary Fig. 1c).

KIKS locus undergoes low-level CSR-like recombination.
To increase the probability of CSR events in spleen, we immunized KIKS mice using intraperitoneal injections of sheep red blood cells (SRBC). Eight days after immunization, mice were analysed. As shown in Supplementary Fig. 1e, germinal centre (GC) generation is similar in KIKS and WT mice, demonstrating that mutant B cells efficiently participate to immune responses. However, o0.2% of B cells expressing only mk-LC were detectable in both KIKS AID þ / þ and AID þ / À mice (Fig. 2a) despite a clear increase of IgG1 and IgG3-switched cells, witnessing parallel efficient CSR at the IgH locus (Fig. 2b,c). These results were confirmed in mice analysed 10 days after booster immunization with SRBC ( Fig. 2c). Finally, weakness of KIKS CSR was emphasized by the minute level of serum mk-LC yielded on immunization as compared with IgG1 and IgG3 (Fig. 2d). We then analysed PP known to abundantly switch to IgA ( Supplementary Fig. 2a). A small but significant proportion of these PP B cells was consistently found to express murine k-LC (0.68 ± 0.09% in KIKS AID þ / þ and 0.63 ± 0.13% in KIKS AID þ / À mice, mean±s.e.m.; Fig. 2c,e), a population specifically enriched among GC B cells (Fig. 2e, lower panel). AID dependence of these CSR-like events was confirmed by their absence in AID-deficient KIKS mice (Fig. 2c,e). Finally, we induced CSR in KIKS B cells by 4 days in vitro stimulation. Cytometry analysis indicated that, despite normal IgG3 and IgG1 CSR, the percentage of mk-LC-expressing cells still remained extremely low (0.37 ± 0.05% and 0.40 ± 0.04%, mean ± s.e.m., after lipopolysaccharide (LPS) or anti-CD40 þ IL-4 stimulation, respectively) albeit significantly higher than in AID À / À stimulated B cells ( Supplementary Fig. 2b). Altogether, these data show that the KIKS locus can undergo CSR-like events but only at a low level and in conditions only achieved on strong B-cell stimulation, such as in mucosae-associated lymphoid tissues.
Knocked-in switch regions are involved in CSR-like events. To further confirm that the few B cells expressing mk-LC carried CSR-like recombination, Sm/Sg3 junctions were cloned and sequenced from in vitro-stimulated B cells (LPS and CD40 þ IL4) and PP cells. Nested PCR primers amplified genomic DNA from the Jk5 exon to the 5 0 part of the loxP site ( Supplementary  Fig. 3a,b), and yielded bands in PP cells from KIKS AID þ / þ and AID þ / À mice but not from KIKS AID À / À cells or stimulated B cells ( Supplementary Fig. 3c). We then cloned the amplified bands, which indeed corresponded to junctions of break points located within or close to Sm and Sg3. Locations of break points are depicted on Fig. 3a and junction sequences are shown in Supplementary Fig. 4. Strikingly, Sm and Sg3 sequences around junctions carried very high mutation loads of 33.5 and 25.3 mutations per kb, respectively (Fig. 3b). Junctions involved essentially direct joins or short microhomologies (r3 nt), with no apparent bias toward A-NHEJ (Fig. 3c). In conclusion, CSR-like events occurred at low level in the KIKS S regions but associated with intense local AID recruitment and SHM levels far above those usually reported around normal IgH S-junctions 17 .
Knocked-in switch regions of KIKS are accessible to AID. Scarcity of KIKS CSR-like recombination could have reflected poor AID recruitment on the knocked-in Sm and/or Sg3. As AID targeting of S regions needs transcription, we checked the transcriptional activity at inserted S regions. It can obviously be inferred from the normal level of hk-LC expression in KIKS B cells that Sm is efficiently transcribed and spliced out. Likewise, we also readily detected LVk4-63/mCk transcripts implying transcription and splicing of the inserted Sg3 ( Supplementary  Fig. 5b). To more specifically address the transcriptional abundance of KIKS S regions, we carried out quantitative PCR on primary transcripts of both KIKS Sm and Sg3. We observed about fivefold less Sg3 transcripts compared with Sm, a difference comparable to that obtained at the IgH locus in LPS-stimulated B cells, a condition known to efficiently induce Ig3 transcription and CSR to IgG3 (ref. 12; Supplementary Fig. 5b). Since both S regions are transcribed in a locus naturally accessible to AID, we checked the mutation frequencies within Sm and Sg3 before CSR. For Sm, we analysed 570 bp downstream Jk5, comprising 220 bp from the Jk intron and 350 bp from the inserted Sm (Fig. 4a). As mutations occur within a range of 1.5 kb after the V promoter, with a rate decreasing with promoter distance and highest in the VJ exon, we checked if this distance also affected SHM in Sm and compared KIKS Sm mutations following rearrangement of either distant (Jk1) or proximal (Jk5) VkJk rearrangement (Fig. 4a). For Sg3, we analysed the first 400 bp (Fig. 4b). As expected, mutation frequencies within the knocked-in Sm and Sg3 were far higher in sorted PNA high Fas þ GC B cells than in resting KIKS AID þ / þ PPs B cells (Fig. 4a,b and Table 1). VJk5-rearranged alleles revealed similar SHM frequencies in Jk5 intron and in Sm (11.19 versus 13.08 mutations per kb). This is consistent with data reported in the equivalent region of the WT Igk locus 13,18 . Surprisingly, VJk1-rearranged KIKS loci also revealed high SHM rates within the knocked-in Sm (7.80 mutations per kb; Table 1) although it lies more than 2 kb downstream of pVk, that is, at a distance normally escaping mutations downstream of rearranged VJ exons 19 (Fig. 4a). This high mutation rate at long distances from the promoter likely involved an increased and/or extended AID recruitment on S sequences containing highly repetitive RGYW motifs, as recently demonstrated for Sg1 (ref. 20). However, a similarly frequent SHM affected the Jk5 intron from the same rearranged loci (8.10 mutations per kb). To exclude any local effect due to the close proximity of Sm, analyses were extended upstream of Jk5 in a region lying from B1.3 to 1.8 kb downstream of pVk (Jk3) thus supposed to mostly escape SHM 19 (primer 9, Supplementary Fig. 3   Chromatin immunoprecipitation (ChIP) analysis also showed the accumulation at KIKS S regions of acetylated histone H3K9 (H3K9ac), a chromatin mark known to be associated with the transcriptional activity and accessibility of S regions allowing AID recruitment and activity ( Supplementary Fig. 6, top panel) 24,25 . Thus, the KIKS S regions are strongly targeted by AID and SHM, although this is clearly not followed with efficient CSR.
DSBs occur in KIKS S regions. Given that KIKS S regions are efficiently targeted by AID, CSR-like events scarcity could be due to either low occurrence of DSBs and/or their inefficient repair. To address this issue, internal S deletions (ISDs) witnessing resolved DSBs were appreciated within KIKS sequences. Interestingly, such deletions were readily detected in both KIKS S regions, at higher rates than in the corresponding endogenous Sm region and Jk5 intron (0.230 and 0.263 ISD per kb in Sm and Sg3 regions, respectively, versus 0.054 in endogenous Sm and 0.037 in non-S regions; Fig. 5a). The length of ISDs ranged from 16 to 160 bp, obviously biased by the preferential PCR amplification of short ISDs. The KIKS locus thus clearly undergoes AID-initiated DSBs that are proportional to SHM but only in G-rich sequences, consistent with the concept that S regions form R-loops targeted by AID on both the template and non-template DNA strands that hence end with staggered DSBs 6,14 . Most of KIKS ISDs display blunt-end junctions or short microhomology and therefore likely result from c-NHEJ repair of DSBs ( Fig. 5b and Supplementary Fig. 7). Accordingly, ChIP experiments conducted at the KIKS locus and compared with IgH locus demonstrated the presence of 53BP1, known to protect DNA ends from resection ( Supplementary Fig. 6,  bottom panel) 26 . Finally, since gH2AX was supposed to be dispensable for intraswitch recombination but required for the chromatin remodelling that facilitates distant S region synapsis 23 , we further searched for gH2AX foci associated with the KIKS locus. Immunofluorescence staining followed by fluorescence in situ hybridization (3D immuno-FISH) simultaneously visualized gH2AX and IgH or Igk loci in GL7 þ sorted PP B cells. gH2AX foci associated with Igk loci at frequencies close to IgH (6.8 versus 8.9%, respectively) confirming the presence of DSBs and their normal initial processing through gH2AX recruitment (Fig. 5c,d).

Discussion
This study addresses the long-standing issue of minimal requirements for CSR versus SHM, two processes sharing several common prerequisites but yielding drastically different genomic and cellular changes. In the IgH locus, S regions susceptible to multiple AID attacks on both DNA strands are mandatory targets for CSR and it is unclear whether other cis-elements of the locus play a role in orienting the processing of AID lesions towards DSBs and/or distant S-S recombination rather than SHM. The aim of our study was thus to check whether two transcribed and spliced S regions located at a position highly recruiting AID, would be sufficient for mediating CSR-like events and being joined at a high level in activated B cells. This issue was addressed in an in vivo model by generating KIKS mice, which carried S regions inserted into the Igk locus naturally targeted by AID, without any other regulatory element from the IgH locus. By designing the knock-in so that CSR-like events would change LC expression, we observed by flow cytometry that, despite their efficient transcription and splicing-out, the inserted S regions underwent only rare recombination and junction events, even in B cells undergoing high-rate CSR in the IgH locus. Since AID levels are relevant to CSR efficiency 27 , we cannot totally rule out the possibility that additional transcribed S regions could dilute AID activity and then diminish CSR efficiency at the KIKS locus. However and contrary to the herein reported observations, such a sink effect would be expected on one hand to impact both KIKS and IgH CSR, and on the other hand to be more severe in AID þ / À mice with half-dose of AID. Moreover, inserted Sg3 and Sm regions were mutated in AID-expressing cells, and even both showed mutation rates far above the endogenous Sm region of B cells undergoing CSR in vitro or in vivo 13,[21][22][23]26,28 . Altogether, this argues against the hypothesis of a sink effect resulting in a local purely quantitative defect of AID in the presence of increased number of target regions. Our data demonstrated that AID and the BER and MMR machineries, all common actors of both SHM and CSR, were highly recruited to the inserted core S regions. A striking finding is the unexpected high mutation rates in both the Jk5 intron and flanking Sm region regardless of their distance from the promoter 19 . While Igk locus enhancers are likely responsible for AID recruitment onto both promoters of the KIKS locus, the extended SHM shows that the inserted S region structure per se modulates the pattern of AID recruitment. These results are reminiscent to the robust AID targeting recently reported in a model in which the core Sg1 region was inserted in the Bcl6 locus 20 . Thus, tethering of AID appears as an intrinsic property of transcribed S regions. However, by contrast to the Bcl6 model where regions flanking the core Sg1 were free of AID targeting 20 , we found a sustained high SHM rate throughout the 2-kb-long Jk region that cannot be fully related to the fact that Igk is a much better AID target than Bcl6 (ref. 29). In the KIKS model, the VJ gene close to the pVk is initially subjected to SHM and it is likely that the distal G-rich S region not only recruits but also maintains AID tethering throughout the entire transcribed Jk region as it certainly processes in the IgH locus to target the whole extent of S regions on distances up to 10 kb (ref. 25).
In current models of CSR, richness of core S regions in RGYW motifs is claimed to allow the generation by AID and UNG of staggered DSBs then recruiting the NHEJ machinery 2 . Accordingly, gH2AX foci are readily detected at KIKS loci together with short ISDs in KIKS S regions at higher rates than in the IgH Sm (likely related to the locally increased SHM), confirming the occurrence of DSBs and their repair with no bias towards A-NHEJ. We also readily detected 53BP1 recruitment in the KIKS S regions, further indicating that DNA ends are likely protected from resection and overwhelmingly repaired by classical NHEJ similar to normal IgH S region DSBs 26 . Altogether, these ISDs demonstrate that, besides high AID targeting indicated by SHM, DSBs are also generated and normally processed into KIKS S regions. Therefore, the poor ability of the KIKS locus to undergo CSR most likely reflects a defect in synapsis between the distant Sm and Sg3 and/or their inefficient recruitment into 'CSR factories' supporting recombination at long distance. This deserves comparison with recent studies showing that distant DSBs induced by high efficiency endonucleases (I-SceI meganuclease or Crispr/Cas9) are efficiently processed through a general cis-joining mechanism needing neither local transcription nor specific chromosomal localization, provided that they are comprised within the 1-Mb spreading range of the gH2AX-dependent DNA damage response 26,28,30 . Although such models interestingly underlined that any cell can efficiently handle distant DSBs in cis, it differed from CSR by multiple features. According to the DNA deamination model, CSR uses no endonuclease directly generating DSBs but rather needs processing of single-strand DNA bubbles appearing around mismatches. Together with the low SHM observed in S regions of GC B cells, it is then tempting to speculate that, by contrast to endonuclease-induced DSBs, the occurrence of two synchronous distant AID-induced DNA breaks should be rare in physiology, needing recruitment in specialized CSR factories to be efficiently joined and repaired. In addition, I-Sce-I-induced DSBs underwent extensive end processing 28 , a condition rarely encountered during AID-induced CSR at the IgH or KIKS loci in which blunt ends are essentially processed through c-NHEJ, thus preserving locus integrity. In fact, AID remodelling of the KIKS locus resembles models of 'failed' CSR-like 53BP1 À / À , gH2AX À / À or AID carboxyl-terminal region mutants, with normal SHM and intraswitch region recombination but only rare CSR junctions between distant S regions 23,26,31 . While such defects of the potential 'CSR factory' components are absent in our model (where normal IgH CSR occurs), the KIKS locus might then simply demonstrate a defective recruitment of the knocked-in S regions into such CSR factories. Whether the Igk locus may intrinsically dictates differential outcome of AID-induced lesions, as it was recently demonstrated when S regions were knocked-in in other loci 20,32 , is another possibility. But the presence of ISDs, few S-S junctions, 53BP1 recruitment and gH2AX foci, proves that beside AIDinduced single-strand mutations, DSBs occur at the KIKS locus but are not appropriately handled for efficient synapsis and junction of distant breaks. We thus propose that beside the general possibility for DSBs to be cis-joined over long distances, optimal CSR requires additional IgH cis-regulatory elements recruiting and synapsing target regions into specific nuclear areas with the functional ability to coordinate and associate distant DSBs and promote long-range recombination. Hereby constraining CSR to the IgH locus would also help prevent side genomic alterations. In this view, it would be relevant to check if the presence of S regions outside of their physiological loci could increase, as previously described in c-Myc 32 , translocation potential and genomic instability in KIKS cells and to what extent.
This work brings new insights in the molecular dissection of CSR. It demonstrates in a locus physiologically targeted by AID, that transcription, splicing, efficient SHM and occurrence of DSBs repairs by c-NHEJ into S regions are not sufficient to trigger CSR. We thus propose that the early events of CSR from cytidine deamination to occurrence of DSBs can affect any G-rich transcribed and spliced genomic position targeted by AID. By contrast, only the IgH locus context provides all the additional conditions for long-range synapsis between AID targets and efficient CSR. A good candidate for assembling the latter conditions is obviously the 3 0 RR, as its deletion in mouse drastically impairs CSR and was also shown to impair IgH synapsis 12 . Long-range interactions between IgH 3 0 RR, Em and I-promoter activated by specific stimulation, was shown to allow the acceptor S region to interact with Sm, then contributing to the stability of S-S synapsis during CSR 33 . Since 3 0 RR deletions globally affected IgH SHM, GLT and CSR, its complex functions can hardly be analysed by knockout experiments. Whether the sole downregulation of GLT or the concomitant lack of synapsispromoting elements explains the lack of CSR in 3 0 RR mutant mice remains to be addressed. The adjunction of the 3 0 RR to the KIKS locus could provide insight into this issue. Others potential regulators of CSR are I-promoters that drive acceptor S region transcription before CSR and could participate in recruiting CSR factors 3 . Finally, we also cannot fully rule out the possibility that the KIKS S regions are too close to each other to allow optimal looping of the chromosome during synapse. While further investigations are needed to decipher the exact contribution of each of these elements, the KIKS strategy already sheds new light on the minimal requirements for efficient CSR, that are clearly beyond transcription, AID recruitment and occurrence of DSBs.

Methods
Generation of KIKS mice. The previously described 12.7 kb BamHI fragment from the Igk locus containing the 5 Jk segments and Ck was used 34 . The 0.8-kb BstEII fragment, located 0.2 kb downstream of Jk5 was eliminated and replaced by a NotI site, then allowing the sequential cloning of the fragments composing the KIKS locus using EagI and NotI compatible sites. Sm region comprises a HindIII 1.9 kb fragment from a plasmid containing the murine JH-Cm region 35 with 1.5 kb of core Sm repeats. A 2-kb fragment of the Sg3 sequence (Genebank accession number M12182) was amplified from C57BL/6 genomic DNA, with the indicated primers ( Supplementary Fig. 3b), and features the entire repetitive region, previously described 36 . Equivalent S regions were previously shown to efficiently support CSR in several studies [37][38][39] . From 5 0 to 3 0 , the following parts were then added: the murine core Sm, a 1.5-kb genomic fragment containing human Eik and Ck, a 0.3-kb fragment containing Vk4-63 promoter and the following leader exon (in which the last codon was replaced by a stop codon to mimic a non-translatable I-exon) and the murine core Sg3. A neomycin-resistance gene (Neo R ) flanked with two loxP sites was inserted downstream Sg3. Upstream of the 5 0 arm, a thymidine kinase (TK) gene was inserted for negative selection against random integration. E14 embryonic stem cells were transfected with linearized vector and selected by addition of G418 and ganciclovir to culture medium. An embryonic stem clone showing homologous recombination of the Igk locus by Southern blot was injected into C57BL/6 blastocysts, and resulting chimeras were mated with C57BL/6 animals. Progeny was then bred with EIIa-CRE mice (a gift from Dr H. Westphal, Bethesda, MD) to obtain a Neo R -deleted KIKS lineage and avoid any transcriptional competition from the Neo R promoter. Mice were first screened by PCR for the presence of the human Ck exon and then for homozygous loss of the WT allele. All the animal strains were bred and maintained in pathogen-free conditions and analysed at 8-16 weeks of age. All the protocols used have been approved by our institutional review board for animal experimentation (Comité Régional d'Ethique de l'Expérimentation Animale du Limousin). Sample size was determined to support statistical analysis while minimizing the number of animals involved. No animal was excluded from the analysis.
Immunization and spleen cell cultures. Mice were immunized by intraperitoneal injection of 200 ml of SRBC and analysed after 8 days. In some cases, mice were booster immunized at day 20 with a second injection of 100 ml of SRBC and analysed 10 days after. Total spleen cells were cultured at a density of 1 Â 10 6 cells ml À 1 in RPMI 1640 supplemented with 10% fetal calf serum either with 50 ng ml À 1 LPS (Sigma-Aldrich, St Quentin Fallavier, France), alone or plus 20 ng ml À 1 IL4 (PeproTech, Rochy Hill, NJ) or IL4 plus 5 mg ml À 1 anti-CD40 (R&D Systems, Minneapolis, MN).
Flow cytometry and cell sorting. Antibodies and reagents used for staining and cell sorting experiments are detailed in Supplementary Table 2. Flow cytometry analyses were done on a BD Pharmingen LSRFortessa cytometer. Data were then analysed with BD FACSDiva software (BD Biosciences, San Jose, CA). Intracellular staining for flow cytometry was done using BD Cytofix/Cytoperm Kit following the manufacturer's instructions (BD Biosciences). For SHM analyses, GC (B220 þ /PNA high /Fas þ ) and resting (B220 þ /PNA low /Fas À ) B cells from PP were sorted. Cell sorting experiments were done on a BD Pharmingen FACS-Vantage cell analyser.
SHM and S-junction analyses. Genomic DNA from total PP cells (S-junctions) or sorted B-cell subsets was extracted using standard proteinase K procedures. All DNA sequences for SHM were amplified using Phusion high-fidelity DNA polymerase (New England BioLabs) with primers indicated in Supplementary  Fig. 3, cloned into pGEM-T Easy vector (Promega, Madison, WI), sequenced with the BigDye Terminator kit on a 3130xl Genetic Analyzer (Applied Biosystems) and analysed using Finch TV software. Mutations were counted for each sequence, then added and divided by the total number of analysed nucleotides to obtain mutation frequency. For junction analysis, DNA was subjected to two rounds of PCR with nested primers (Supplementary Fig. 3). The first round of PCR consisted in 10 cycles followed by a second round of 32 cycles.
ELISA assays. Human and mouse k light chains, IgG1 and IgG3 in sera from WT and KIKS mice were evaluated by ELISA (enzyme-linked immunosorbent assay) as previously described 40 . In brief, 96 multiwell plates (Maxisorb, Nunc, Denmark) were coated overnight with 1 mg ml À 1 with appropriate antibodies (Supplementary Table 2) in NaHCO 3 buffer. Plates were then blocked with PBS containing 3% bovine serum albumin for 30 min and diluted sera samples or standards were incubated for 2 h at 37°C. Alkaline phosphatase-conjugated antibodies were incubated for 1 h 30 at 37°C and followed by the addition of p-Nitrophenyl phosphate (Sigma-Aldrich). Reading of plates was done at 405 nm with a SAFAS Xenius device.
Transcript analysis. Total RNA from PP was extracted with TRI Reagent (Ambion, Austin, TX), according to the manufacturer's protocol, and 1 mg was used for cDNA synthesis using hexameres (Applied Biosystems). End-point PCR of mature transcripts was performed using Q-biogen Taq polymerase (Qiagen). Relative quantification of primary transcripts was performed with SYBR green Master Mix (Applied Biosystems) and compared with GADPH levels. Primers are indicated in Supplementary Fig. 3.
Immuno-FISH 3D. GC B cells (GL7 þ ) from murine PP were dropped onto poly-Llysine slides and fixed with 2% paraformaldehyde for 10 min at room temperature. After washing with PBS, cells were permeabilized with 1% Triton-X-100 for 15 min at room temperature. Cells were stained with g-H2AX antibody (1/300, Millipore) overnight at 4°C followed by secondary goat anti mouse IgG (1/1000, Invitrogen) during 1 h at room temperature. After three washes with PBS/Tween 0.1%, cells were postfixed in 2% paraformaldehyde for 10 min at room temperature and permeabilized with Triton-X-100 1% HCl 0.1 M for 20 min at room temperature. Denaturation was performed during 35 min with 2 M HCl at room temperature and slides were rinsed three times in cold PBS. Cells were hybridized overnight at 37°C with IgH (5 0 IgH) or Igk (RP23-341D5) probes labelled with dUTP-digoxigenine (Roche). The next day, cells were washed in 1 Â saline sodium citrate (SSC) at 72°C for 5 min and 1 Â SSC at room temperature for 5 min. Then, the cells were incubated with anti-digoxigenine FITC (1/300, Abcam) 1 h at room temperature and washed three times in 2 Â SSC. Finally, the slides were mounted with vectashield (Vector labs). Images were acquired with an epifluorescence inverted microscope (LEICA DMI6000B). Optical sections separated by 0.2 mm were collected and stacks were deconvoluted using Huygens software and analysed with Image J software.
Chromatin immunoprecipitation. We used the ChIP-IT Express Kit (Active Motif) according to the manufacturer's instructions. In brief, 5 Â 10 6 PP cells were fixed with 1% formaldehyde at room temperature. The reaction was stopped by adding glycine to a final concentration of 0.125 M. After the crosslinking reaction, cell lysis and sonication were performed on Vibracell sonicator (Sonics and Materials, Inc). For immunoprecipitation, 4 mg of each antibody was added to the lysate at 4°C overnight. The immunoprecipitated DNA was used for real-time PCR using SYBR Green Master Mix (Applied Biosystems). The quantitative PCR signal was normalized using the % input method. Primers for KIKS locus are listed in Supplementary Fig. 3 (or as in Wang et al. 25 for the IgH locus). Antibodies used for pull down of DNA associated with H3K9Ac and 53BP1 or negative control antibody (Millipore) are indicated in the Supplementary Table 2 and were previously described 25,41 .