Localized DNA Demethylation at Recombination Intermediates during Immunoglobulin Heavy Chain Gene Assembly

The dynamics of DNA methylation during the complex genomic rearrangement of antigen receptor genes in developing B lymphocytes reveal localized demethylation of the first recombination product that may serve as a mark necessary for the second step of rearrangement.


Introduction
Tissue-specific gene expression requires multiple epigenetic changes. These include nuclear location, chromatin remodeling, covalent histone modifications, and DNA methylation [1][2][3][4]. Recent genome-wide analyses reveal several important correlations between gene activity and epigenetic modifications. Transcriptionally inactive genes are marked by histone H3 methylation at lysines 9 and 27 (H3K9me, H3K27me), whereas histone acetylation and H3K4 methylation are associated with gene activity [5]. H3K9 acetylation and H3K4 methylation have been inversely correlated with CpG methylation, consistent with the long-established view that transcriptionally active promoters are hypomethylated [6].
Most studies of CpG methylation vis-à-vis transcription have focused on gene promoters, particularly those that contain CpG islands (CGI). Such promoters are typically un-methylated and are transcriptionally active; conversely, methylated CGI promoters are usually transcriptionally silent. The role of CpG methylation in non-CGI promoters is less clear [7]. Many tissue-specific promoters fall in this category and, in a limited number of cases where examined, such promoters also have reduced CpG methylation in tissues where they are active. Most recently, the advent of whole genome CpG mapping has drawn attention to possible functions of CpG methylation within gene bodies [8,9]. Proposed functions for such CpGs include transcription elongation and the regulation of splicing. The lack of a coherent picture for the role of CpG methylation in non-CGI contexts indicates an ongoing need for analysis of CpG methylation, particularly in tissue-specific genes that lack CGIs.
Antigen receptor genes of B and T lymphocytes serve as excellent paradigms for developmentally regulated gene expression. Immunoglobulin heavy chain (IgH) genes are assembled during B lymphocyte development by juxtaposition of variable (V H ), diversity (D H ), and joining (J H ) gene segments by a process known as V(D)J recombination [10]. The mouse genome contains approximately 150 V H gene segments, 10 to 13 D H gene segments, and four J H gene segments, dispersed over 2 Mb [11,12]. D H to J H recombination occurs first, followed by V H recombination to the rearranged DJ H junction ( Figure 1). These two steps produce a fully recombined V(D)J allele, with the potential to encode heavy chain protein. IgH expressing cells develop into pre-B cells where immunoglobulin k or l light chain genes rearrange. Cells that make both IgH and IgL polypeptides express cell surface immunoglobulin, and are exported out of the bone marrow to become mature functional B cells.
DNA methylation was the earliest epigenetic mark associated with immunoglobulin gene regulation [13,14]. Using a variety of cell lines, these pioneering studies showed that Vk promoters that were rearranged, and therefore transcriptionally active, lacked CpG methylation. Because these studies used restriction enzyme isoschisomers, it was difficult to define the extent of demethylation on recombined k alleles in these studies; however, a site located approximately 10 kb 59 of the promoter of a rearranged Vk21 remained methylated. Subsequently, studies by Bergman, Cedar and colleagues showed that one k allele is preferentially demethylated at a site close to the Ck exons at the pre-B cell stage of B cell differentiation. Preferential Vk recombination of the demethylated allele led them to propose that demethylation permits Vk to Jk recombination [15]. Analyses of transgenic mice containing synthetic recombination substitutes were also consistent with the idea that demethylated alleles recombine preferentially [16,17]. Additionally, de-methylation state of this site has also been proposed to play a role in somatic hypermutation in germinal centers [18].
Compared to these extensive studies at the k locus, relatively little is known about developmental regulation of DNA methylation at the IgH locus. Two relatively recent studies evaluated the role of methylation in V H gene rearrangements [19,20]. Within the small family of V H S107 genes, the methylation status of one CpG in pro-B cells correlated with recombination potential. Both studies also examined the status of the 39-most V H gene segment, V H 81X, that rearranges most frequently. Two CpG sites within the coding region remained methylated in pro-B cells whereas a CpG near the recombination signal sequence was demethylated. However, the extent of demethylation of all genes examined was comparable in pro-B cells where V H genes recombine and in non-B lineage cells where V H genes do not recombine. Thus, methylation status could not be directly correlated with rearrangement potential. More importantly, virtually nothing is known about CpG methylation in D H -Cm part of the locus, where V(D)J recombination is initiated [21].
Here we used the bisulfite modification method to assay DNA methylation in the D H -Cm domain of the IgH locus prior to the onset of recombination and as recombination proceeds. The only sites of extensive tissue-specific DNA demethylation in the unrearranged locus corresponded to a promoter close to DQ52 and the intronic enhancer Em. Additionally, sequences close to the intergenic control region [22], which marks the 59 boundary of the D H -Cm domain, were partially demethylated. These regions correspond to the only known DNase1 hypersensitive sites in this part of the IgH locus. All CpGs close to, or within, other rearrangeable D H or J H gene segments remained hypermethylated regardless of whether they were marked with activation or inactivation histone modifications. However, DJ H junctions that were generated after the first recombination step were completely demethylated. Junctional demethylation was highly localized, B lineage-specific, and required the presence of an intact Em. These observations suggest that localized demethylation of the DJ H junction occurs after the first recombination step and may mark the junction for subsequent V H gene rearrangements.

Tissue-Specific DNA Demethylation Coincides with DNase I Hypersensitive Sites
Histone modifications across the unrearranged D H -Cm domain of the IgH locus show marked heterogeneity. The region encompassing the J H gene segments is marked by the highest levels of histone acetylation and histone H3 trimethylation at lysine 4 (H3K4me3) (Figure 1) [23][24][25]. The 59-and 39 D H gene segments, DFL16.1 and DQ52, are also associated with these active modifications, whereas the six to nine DSP gene segments that lie in between are marked by repressive H3 dimethylation at lysine 9 (H3K9me2). Two tissue-specific DNase l hypersensitive sites (DHS) in this part of the locus correspond to a promoter associated with DQ52 and the intronic enhancer Em. A cluster of three DHSs was recently identified just 59 of DFL16.1 [22]. These latter regions contain binding sites for the transcription factor CTCF [26] and have been proposed to serve as the 59 boundary of the D H -Cm domain [27].
To investigate the DNA methylation state of the unrearranged locus, we purified genomic DNA from pro-B cells obtained from the bone marrow of RAG2-deficient mice to use in bisulfite modification assays modified from Frommer et al. [28]. We used at least two independent DNA preparations starting with cells obtained from six mice for each sample. Methylation profiles of the regions amplified from two independent experiments were comparable ( Figure S1). We sequenced approximately 12-50 alleles (Table S1), of which 12 representative clones are shown throughout the text. The mb-1 promoter and b-globin locus control region served as positive [29] and negative controls [30], respectively ( Figure S2); we used CD4 + CD8 + double positive thymocytes (DP) as a lineage control and kidney genomic DNA as a non-lymphoid control. The relative positions of all CpGs are shown in Figure S3.
We found that sequences near DQ52 and Em were substantially demethylated in RAG2-deficient pro-B cell DNA compared to kidney DNA ( Figure 2). In contrast, unrearranged DFL16.1 and DSP gene segments, which are marked by active and inactive histone modifications [23], respectively, were methylated. While the methylation status of DSP gene segments was consistent with earlier association of H3K9me2 modification with CpG methyl-

Author Summary
DNA methylation at CpG dinucleotides is implicated in the regulation of gene expression in mammals. However, the regulation of DNA methylation itself is less clear despite recent advances in identifying enzymes that add or remove methyl groups. We have investigated the dynamics of DNA methylation during genome rearrangements that assemble antigen receptor genes in developing B lymphocytes to determine whether methylation status correlates with rearrangement potential. Two recombination events generate immunoglobulin heavy chain (IgH) genes. The first step brings together diversity (D H ) and joining (J H ) gene segments to produce DJ H junctions. We show that both gene segments are methylated prior to rearrangement, whereas the DJ H product is demethylated. DJ H junctional demethylation is tissue-specific and requires an enhancer, Em, located within the IgH locus. The latter observations indicate that localized demethylation of the DJ H junction occurs after the first recombination step and thus does not guide this first step of IgH gene assembly. Our working hypothesis is that recombination induces demethylation of recombinant product and may mark the junction for the second step of IgH rearrangement, juxtaposition of variable (V H ) gene segments to rearranged DJ H products to produce fully recombined V(D)J alleles.
ation [31], the methylated state of DFL16.1 did not follow this paradigm. An amplicon that included the J H 1 gene segment located within the peak of active modifications was partially demethylated in pro-B cells compared to kidney. Note that the DQ52 and J H 1 amplicons, that are demethylated and partially methylated, respectively, are separated by only about 170 nucleotides that contain two CpGs. These observations indicate that the prominent peak of H3ac and H3K4me3 encompassing the J H gene segments is not sufficient to direct CpG demethylation of this region. Moreover, the J H region has been shown to bind the highest levels of recombinase proteins RAG1 and 2, to form a recombination center at which IgH rearrangement initiates [21]. Clearly, the partially methylated state of this region does not preclude RAG 1/2 recruitment. Additionally, DQ52 and  [11,12]. Exons encoding the constant regions of IgM and IgD are indicated as Cm and Cd. A promoter 59 of DQ52, the 39-most D H gene segment, is indicated by the yellow oval and the intronic enhancer Em by a teal oval. Top line shows the germline (GL) configuration with associated histone modifications in B lineage precursors [23][24][25]. Histone H3 and H4 acetylation are shown in orange and presence of heterochromatic H3K9 methylation by the red line. Vertical red arrows represent the tissue-specific DNase I hypersensitive sites in the germline state. Next two lines show sequential stages of VDJ recombination at the IgH locus. D H to J H rearrangement occurs first resulting in a DJ H junction and, depending on which D H rearranges, residual upstream unrearranged D H s may be present. V H rearrangement occurs to the DJ H junction to generate a VDJ H junction; during this process unrearranged D H s are lost from the genome. doi:10.1371/journal.pbio.1001475.g001  [32][33][34][35], yet they had distinctly different methylation states. The methylation status of D H and J H gene segments suggests that the first step of Ig gene assembly may not be regulated by DNA methylation.
Because both DQ52 and Em coincide with tissue-specific DHSs, we examined the methylation status of newly identified DHSs located approximately 6, 4, and 1 kb 59 of DFL16.1. We found two clusters of CpG dinucleotides centered approximately at 6.5 and 4 kb 59 of DFL16.1. The precise location of these regions and the relative location of CpG dinucleotides within each region are shown in Figure S3. The cluster at 26.5 was partially demethylated in primary pro-B cells compared to DP thymocytes that served as a non-B lineage control ( Figure 3). The cluster at 24 kb was similarly methylated in both cell types. Because neither location corresponded precisely with the DHSs, we also examined shorter stretches of CpG dinucleotides in this region. Two sites located at 26 kb were significantly demethylated in pro-B cells compared to DP thymocytes, whereas one out of three CpGs at 25 kb was completely demethylated in pro-B cells. Two CpGs at 23 kb were completely demethylated in both pro-B cells and DP thymocytes. Finally, a cluster of CpGs in the 21.3 kb region was largely methylated in pro-B cells (see below). Overall, the majority of sites in this region remained methylated in pro-B cells, with very specific sites being targeted for demethylation. Possible interpretations of these observations are discussed below.

Localized DNA Methylation at Recombination Intermediates
To determine whether DNA methylation was altered by recombination, we examined pro-B cells purified from the bone marrow of wild-type mice, which contain DJ H junctions as well as unrearranged alleles. We found that unrearranged DFL16.1, DSP, and J H regions were methylated, while Em region was demethylated ( Figure 4A) as seen in RAG2-deficient pro-B cells (note that un-rearranged DFL16.1 and DSP regions in these cells represent fully germline IgH alleles as well as those in which a downstream D H gene segment has undergone recombination to form a DJ H junction). However, DFL16.1/J H 1 and DSP/J H 1 recombined junctions were demethylated ( Figures 4B and S4). For junctions that involved DFL16.1, demethylation of both parts of the recombination reaction was evident due to the presence of analyzable CpGs in each segment. Many DSP-involving junctions lacked CpG dinucleotides contributed from the DSP part. This is because all DSP gene segments lie within a 4-kb repeat unit [23,36], which is reflected in all DSP family members sharing a common CpG located 39 of the gene segment. Two DSP gene segments, DSP2.2 and DSP2.3, have a CpG within the gene segment. In addition, the amplicon encompassing DSP2.3, includes a third CpG dinucleotide 59 of the gene segment ( Figure  S3). The 39 CpG is lost during recombination, and deletions that occur during recombination lead to many junctions that have no CpG contribution from the D H part ( Figures 4B, 4C, and S4). Yet, demethylation of DSP/J H junctions was evident from the state of residual CpGs that fell within the J H 1 region.
During B cell development, pro-B cells that contain a functional IgH allele undergo several rounds of cell division and differentiate into small pre-B cells [37]. We found that pre-B cells also contained demethylated DFL16.1/J H 1 and DSP/J H 1 junctions ( Figure 4C). Sequence diversity at recombination junctions provided an unequivocal measure of allelic individuality in our experiments. For example, many completely demethylated alleles, such as DFL16.1/J H 1 junctions in pre-B cells, could have arisen by aberrant amplification of a single allele. However, analysis of the junctional sequences ( Figure S4) showed that each of these represented unique alleles.
Demethylated DJ H junctions could arise by preferential juxtaposition of demethylated D H and J H gene segments; alternatively, demethylation could be targeted to the junction after recombination. One prediction of allelic demethylation prior to recombination is that unrearranged D H s located 59 of the DJ H junction (see Figure 1) should be demethylated. To determine whether DNA demethylation was specific to the DJ H junction, or occurred more widely over recombinant allele, we used mature B cells. Earlier studies have shown that about 40% of mature B cells from wild-type mice contain VDJ H rearrangement on both alleles [37,38]. In such cells, unrearranged D H gene segments are lost from the genome. The remainder contain one VDJ H rearranged allele and one DJ H rearranged allele; these cells therefore contain unrearranged D H gene segments 59 to the rearranged DJ H junction ( Figure 5A). Mature B cells with germline IgH alleles have not been identified in normal mice. First, we determined the methylation state of the DJ H junctions in mature B cells. As seen in pro-or pre-B cells, DFL16.1/J H 1 and DSP/J H 1 junctions were demethylated in mature B cells ( Figure 5C). We then analyzed germline DFL16.1 in mature B cells to score for all alleles with downstream D H rearrangements. We also analyzed a region approximately 1.3 kb 59 of DFL16.1 that would be intact on DFL16.1 rearranged alleles. We found that both these regions were hypermethylated in mature B cells approximating that seen in RAG2-deficient pro-B cells ( Figure 5B). We conclude that demethylation of recombined alleles is highly localized to the DJ H junction.

DNA Demethylation Requires the Intronic Enhancer Em
To understand the basis for localized demethylation, we considered the following possibilities. First, a specific D H gene segment and the J H region could be demethylated immediately before recombination. In this model demethylation would precede rearrangement and could possibly target the recombinase to a preselected D H gene segment. Second, demethylation may occur during recombination, after the DNA breaks have been introduced by recombinase. Third, demethylation of the DJ H junction could occur after rearrangement. The last model predicts the existence of methylated DJ H junctions that are somehow trapped between recombination and demethylation steps. As described below, we found two sources of such DJ H recombined alleles.
The intronic enhancer Em controls key features of the IgH locus assembly via its effects on chromatin structure. Of note, Emdeleted IgH alleles undergo D H recombination but not V H allele and each vertical line methylation of a specific cytosine residue. J H 1-associated CpGs are indicated as squares. Filled and open circles, or squares, indicate methylated and unmethylated cytosines, respectively. Methylation at each position is summarized in the form of pie charts with the percentage of methylated residues indicated numerically. Asterisks replace pie charts where the number of sequences falls below 12. Data shown are derived from at least two independent preparations of pooled bone marrow cells obtained from four to six mice each. Kidney genomic DNA served as the non-B lineage control and the previously characterized mb-1 promoter served as a B lineage-specific positive control. Primers for the DSP region amplify all DSP gene segments from the repeat array. Because the numbers of CpGs vary between DSP gene segments, three different CpG patterns are evident in this amplicon. doi:10.1371/journal.pbio.1001475.g002    [39,40]. We have recently found that the Em enhancer directs localized histone modification changes at DJ H junctions and proposed that these changes configure the IgH locus for V H gene recombination [41]. To determine whether Em affected the methylation status of the DJ H junction, we applied the bisulfite procedure to genomic DNA of bone marrow pro-B cells from mice that lack a 220 nucleotide Em core [40]. We first examined the two regulatory regions. Three of five CpG sites used to assess Em DNA methylation flank the core enhancer and are therefore present on the deleted alleles. These three sites were highly methylated on Em-deficient alleles ( Figure 6A). CpGs associated with the juxtaposed loxP sites that substitute the Em core were also hypermethylated ( Figure S5). Thus, a functional enhancer is required to demethylate this region. The J H 1 region remained hypermethylated in Em 2 alleles, whereas the DQ52 region was unmethylated on Em 2 alleles indicating that its demethylation was Em-independent ( Figure 6A).
In striking contrast to our observations on Em-sufficient alleles, we found that DFL16.1/J H 1and DSP/J H 1 junctions were hypermethylated in Em-deficient pro-B cells (compare Figure 6B to Figures 4B, 4C, and 5C). Moreover, non-templated CpGs introduced by N region addition during recombination [37] were also methylated ( Figure 6B, indicated by black diamonds). Interestingly, DQ52/J H 1 junctions were of a hybrid nature, consisting of several demethylated CpGs from DQ52 and several methylated CpGs from the J H 1 portions of the junction. We infer that demethylation at DJ H junctions requires Em. The state of DQ52/J H 1 junctions indicated that neither rearranging segment imposed its pattern of methylation during the ''cut-and-paste'' reaction of V(D)J recombination. These observations are consistent with a model where active demethylation follows recombination.

Junctional Demethylation Is Tissue Specific
D H to J H recombination does not occur exclusively in B lymphocyte precursors. In particular, DJ H junctions have been detected in a significant proportion of DP thymocytes [42,43]. To determine whether DJ H junction demethylation was lineage restricted, we analyzed DNA from DP thymocytes from wild-type mice. We found that DQ52 and Em regions were demethylated, and germline DFL16.1 and J H regions were hypermethylated as seen in pro-B cells ( Figure 7A). However, both DFL16.1/J H 1 and DSP/J H 1 junctions were considerably more methylated in DP cells than their state in pro-B cells ( Figure 7B), and DQ52/J H 1 junctions displayed the hybrid phenotype seen in Em-deficient pro-B cells. In these cases of methylated DJ H junctions we again observed de novo methylation of non-templated N region CpGs ( Figure 7B, indicated by black diamonds). Thus, despite the presence of an intact demethylated enhancer, DJ H junctions in DP cells retained the methylation status of the rearranging gene segments. Our interpretation is that D H to J H recombination does not require prior DNA demethylation. After rearrangement, Em activity drives demethylation in pro-B cells; however, the milieu in DP thymocytes in terms of Em activity or availability of transcription factors, does not permit efficient junctional DJ H demethylation.

Discussion
Systematic analysis of DNA methylation of unrearranged and partially rearranged IgH alleles in primary cells revealed several interesting features. First, the lack of correlation between histone modifications and DNA methylation status indicates that these marks are independently regulated. This was most clearly evident in the J H region that is marked with the highest levels of histone acetylation and H3K4me3 in the unrearranged state, yet remains hypermethylated. Furthermore, germline DFL16.1 and the DSP gene segments were comparably methylated despite being differentially marked at the level of histone modifications. These observations appear to be at odds with genomic studies that show a correlation between histone acetylation and DNA demethylation, and H3K9 methylation and DNA methylation [31,44,45]. We suggest that the results from genome-wide studies may be skewed towards promoter-based CGIs. Instead, the sites we analyzed were largely non-CGIs and included enhancers (such as Em), promoters (such as DQ52), cryptic promoters (such as DSP and DFL16.1, see below), and regions that cannot be categorized as any of these.
An important corollary of this observation is that prerearrangement DNA methylation status correlates poorly with recombination potential. Recent studies identified the J H region as a RAG1/2-rich recombination center [21]. Our observation suggests that DNA demethylation is not required to generate the recombination center. Additionally, DFL16.1 and DQ52 gene segments that rearrange most frequently [32][33][34][35] have very different levels of DNA methylation prior to rearrangement; conversely, DSP gene segments that rearrange relatively less frequently have comparable levels of DNA methylation as DFL16.1. Taken together, our working model is that DNA methylation status does not guide the first step of IgH gene assembly.
We note, however, that we cannot unequivocally rule out the possibility that selective demethylation of specific D H and J H segments occur in a subset of cells just prior to recombination. For example, while the J H 1 region that we analyzed was largely methylated in pro-B cells prior to rearrangement, we observed that a minority of alleles were demethylated in both RAG-sufficient and RAG-deficient pro-B cells. These demethylated J H 1 alleles may represent a subset of cells in which D H recombination will occur preferentially to the J H 1 gene segment. It will be interesting to determine whether other J H gene segments are similarly singled out for demethylation in subsets of pro-B cells.
Second, sites of maximal tissue-specific CpG demethylation in the germline IgH locus corresponded to the two strongest DNase 1 hypersensitive sites in the 70-kb D H -Cm domain, DQ52 and Em. Because these regions share transcriptional regulatory properties, it is likely that their demethylation is linked to transcription or transcription-associated chromatin changes. Additionally, we noted highly selective CpG demethylation of some sequences and partial demethylation of others 59 of DFL16.1 where three additional DNase 1 HSs have been recently identified. Though these observations link DNase 1 hypersensitivity to DNA methylation, this doesn't seem to be always the case. For example, J H region is known to be highly sensitive to DNase 1 digestion [46] (without being associated with ''classical'' DNase 1 hypersensitive sites), yet it is not demethylated. A possible model for this pattern of methylation emerges from the recent demonstration that Em is Figure 6. DNA methylation state of IgH alleles in the absence of the intronic enhancer Em. Pro-B cells were purified from the bone marrow of Em-deficient mice [40]. Genomic DNA was subjected to bisulfite modification followed by PCR amplification of germline gene segments (A) and DJ H junctions (B). The 220-bp deletion of Em is represented as partial blue arcs separated by a gap. Filled and open circles, or squares, indicate methylated and unmethylated cytosines, respectively. Numbers within regions marked as DFL16.1, DSP, and J H 1 denote CpG dinucleotides corresponding to the configuration at the respective unrearranged gene segments. For example, of the five CpGs at unrearranged DFL16.1 only the first two are retained in DFL16.1/J H 1 junctions. Three out of five cytosines used to analyze wild-type alleles remain in this deletion, and their methylation status is shown immediately below the disrupted enhancer. Sequences that substitute for the enhancer also contain CpGs whose methylation status is shown in Figure S5. (B) Recombined DFL16.1, DSP and DQ52 to J H 1 junctions were amplified, cloned, and sequenced. Untemplated CpGs incorporated during VDJ recombination were found to be methylated (filled diamonds). Data shown are derived from two independent preparations of pro-B cells from Em-deleted mice, with four to six mice in each experiment. doi:10.1371/journal.pbio. 1001475.g006 in spatial proximity to the cluster of DHSs 59 of DFL16.1 [27,47,48]. Since Em itself has a demethylating potential (Em flanking regions are methylated on Em-deficient alleles), perhaps Em looping to 59 DFL16.1 sites ''delivers'' demethylating activities to this region. The partial demethylation observed at 59 DFL16.1 sites may reflect that such loops are present in only a fraction of pro-B cells at any given time. Consistent with the idea that Em induces demethylation in a subset of cells, the form of partial Figure 7. DNA methylation of IgH alleles in thymocytes. CD4 + CD8 + thymocytes from C57BL/6 mice were enriched by adsorption to PNAcoated plastic plates. Genomic DNA purified from these cells was used in bisulfite modification analysis. Amplicons corresponding to unrearranged parts of IgH allele (A) and DJ H junctions (B) were cloned and sequenced. Circles and squares depict CpGs corresponding to D H and J H 1 cytosines, respectively. Filled circles/squares correspond to methylated cytosines; pie charts summarize the percentage of methylated cytosines at each position except where the number of sequenced alleles falls below 12, indicated by asterisks. Untemplated CpGs incorporated during VDJ recombination were found to be methylated (filled diamonds). Data shown were obtained from two independent preparations of thymocytes, using one mouse per experiment. doi:10.1371/journal.pbio.1001475.g007 demethylation seen at DFL16.1 (26.5) involves extensive demethylation of some alleles and essentially complete methylation of others (Figure 3). Em has also been proposed to loop to DHS5-7 of the 39 regulatory region at the 39 of the IgH locus. Interestingly, earlier studies by Giambra et al. [49] showed that DHS5-7 was partially demethylated in a pro-B cell line in a manner similar to the methylation state 59 of DFL16.1. This pattern may be the consequence of Em looping to DHS5-7 in a subset of pro-B cells. Overall, we propose that the pattern of DNA demethylation in the germline IgH locus is determined by spatial proximity to Em.
Third, we found that DJ H junctions were extensively demethylated. Though we cannot discount the possibility that germline D H and J H gene segments were demethylated just before recombination, we favor the hypothesis that demethylation of DJ H junction occurred after recombination. The reasons for this are two-fold: first, DJ H junctions remain methylated in Em-deficient pro-B cells and in CD4 + CD8 + thymocytes. Our interpretation is that these DJ H alleles are caught in an intermediate stage where recombination has occurred but demethylation has not. Second, we think it is unlikely that demethylation occurred prior to recombination followed by re-methylation of DJ H junctions based on the state of DQ52/J H 1 junctions. In both Emdeficient pro-B cells and DP thymocytes the methylation status of DQ52/J H 1 junctions was such that the DQ52 portion was demethylated and the J H 1 portion was methylated. If demethylation preceded rearrangement, generation of each hybrid junction would require remethylation of a subset of closely associated CpGs.
Though our data clearly demonstrate that DNA demethylation of associated gene segments is not essential for D H to J H recombination, we note that the two instances of methylated DJ H junctions identified in this study involve circumstances where the frequency of D H recombination is lower than in wild-type pro-B cells. D H recombination has been estimated to be 5-to 10-fold lower in Em-deleted pro-B cells [39] and only about 30%-50% of IgH alleles in DP thymocytes [42] have DJ H junctions. We cannot rule out, therefore, that gene segment demethylation may increase the efficiency of D H recombination in wild-type pro-B cells. Because Em deletion also leads to loss of other accessibility-associated epigenetic marks in the unrearranged locus, it is difficult to deconvolute the contribution of each mark to recombination efficiency.
How might DNA demethylation be targeted to DJ H junctions? We have previously shown that D H recombination activates cryptic bi-directional promoters associated with most D H gene segments [23]. Transcriptional activity and associated RNA polymerase II recruitment is restricted to the DJ H junctions, and drops off before the unrearranged D H segments located 59 of the junction. These changes require Em and the simplest interpretation is that recombination places cryptic D H promoters under the influence of Em. Our working model is that D H -promoter/Em interaction brings Em-associated demethylating activity to DJ H junctions, thereby resulting in Em-directed demethylation. Because Em's influence does not extend to the next upstream D H gene segment, DJ H demethylation is highly localized. In this model the substantial demethylation of DJ H junctions compared to partial demethylation of looping sites may be due to stronger interaction of Em with DJ H promoters compared to Em interaction with the 39 RR or 59 of DFL16.1 sequences.
Does DJ H demethylation serve a function? In this regard it is interesting to note that V H recombination is significantly reduced in both instances where DJ H junctions remain methylated (Em-deleted alleles and in DP thymocytes). Though this is consistent with the view that DJ H demethylation facilitates V H recombination, we think that the regulation of V H recombination is more complex. The highly localized Em-dependent DJ H demethylation that we describe in this report adds to the emerging evidence that DJ H junctions are distinguished from un-rearranged D H gene segments by several forms of epigenetic changes [41]. These include activation-associated histone modifications, such as H3 acetylation and H3K4me3 and increased sensitivity to DNase I. Like DJ H demethylation, these alterations are also restricted to DJ H junctions and are Em dependent. Taken together, our working model is that all these DJ H -restricted epigenetic changes work in concert to promote the timing and precision of V H recombination.

Mice
Rag2-deficient and C57BL/6 mice were obtained from Jackson Laboratory (Bar Harbor, Maine) and housed in pathogen-free facilities at the NIA or at the University of Massachusetts Medical School. Em-deficient mice were generated and maintained in the facilities at Childrens Hospital (Boston, Massachusetts). Mouse experiments were approved by the Animal Care and Use Committees at the NIA, Harvard Medical School, and the University of Massachusetts Medical School.

Cell Purification
A magnetic cell-sorting system was used to purify pro-B and mature B cells as previously described [25]. Briefly, to obtain pro-B cells with IgH in germline configuration bone marrow cells were recovered from 6-8-wk-old RAG2-deficient mice by flushing the femur and tibia with 10% calf serum in PBS. CD19+ cells were purified from single cell suspension of bone marrow by positive selection using paramagnetic microbeads (Miltenyi Biotech). To obtain mature B cells, the same procedure was followed using cells obtained from the spleen of C57BL/6 mice. The resulting cells were greater than 90% CD19+ by flow cytometry.

FACS Sorting of Pre-B Cells and Pro-B cells from C57BL/6 Mice
Freshly isolated bone marrow cells were re-suspended to 6610 7 /ml in staining media containing biotin-, flavin-, and phenol red-deficient RPMI 1640 (Irvine Scientific), 10 mM Hepes, pH 7.2, 0.02% sodium azide, 1 mM EDTA, and 3% newborn calf serum, and treated with Fc block for 10 min on ice. Cells were incubated with primary antibodies for 20 min and then washed three times, incubated with SA-Cy5PE for 15 min, and then washed three more times. After the final wash, samples were resuspended in 1 mg/ml propidium iodide to exclude dead cells. Primary antibodies included B220 APC; CD43 (clone S7) PE; IgM FITC; Ly6C FITC; BP-1 (clone 6C3) biotin. Antibodies were purchased from BD Biosciences/Pharmingen, eBioscience, or CALTAG. FACS Sorting was performed on a 3-laser, 9-detector MoFlo. Data were analyzed for presentation purposes with FlowJo software (Tree Star).

Genomic DNA Isolation and Modification
Genomic DNA was extracted from 1-5610 6 cells by using the DNeasy Blood & Tissue Kit (QIAGEN) according to the manufacturer's protocol.
Sodium bisulfite modification was carried out as described earlier [28] with some modifications. Briefly, 1 mg of genomic DNA was denatured by incubating with NaOH at 37uC for 15 min. Sodium metabisulfite (3 M) and hydriquinone (0.5 mM) were added to the samples together and incubated overnight at 55uC. The reaction was stopped by incubating samples with NaOH 15 min at 37uC. Modified DNA was purified using GeneClean II (MP Biomedicals). Modified DNA was precipitated and dissolved in 30 ml of 1 mM Tris-HCl (pH 8.0).
Modified templates were amplified by nested PCR using primers listed in Table S2. PCR products were separated by agarose gel electrophoresis, purified using QIAquick gel extraction kit (QIAGEN), and cloned into pCRII-TOPO vector. Mini prep DNA containing amplicon inserts were identified by PCR and sequenced commercially (SeqWright). Sequence analysis showed 99%-100% bisulfite modification efficiency ( Figure S6). Figure S1 Comparison of DNA methylation status between two independent experiments. For each region analyzed in the paper, we used at least two independent DNA preparations starting with cells obtained from six mice for each sample. Methylation profiles of the regions amplified from two independent experiments were comparable, as shown with the example of the DFL16.1 region in RAG 2/2 pro-B mice. (TIF) Figure S2 DNA methylation state of the mb-1 and bglobin locus activating region in different cell types. For our experiments we have chosen the previously characterized mb-1 gene and its promoter as a positive and b-globin locus activating region as a negative control. Sequences from at least 12 individual colonies were analyzed for each cell type. Consistent with the observations of the Hagman group [29] mb-1 gene and its promoter are hypomethylated in early stages of B cell development. As expected mb-1 and its promoter were methylated in CD4 + CD8 + DP thymocytes and kidney cells, which served as a negative control for this region. b-globin locus activating region was methylated in a RAG2-deficient pro-B cell line, primary RAG2-deficient pro-B cells, and kidney.