Noncoding RNA transcription targets AID to divergently transcribed loci in B cells

Abstract

The vast majority of the mammalian genome has the potential to expressnoncoding RNA (ncRNA). The 11-subunit RNA exosome complex is the main source of cellular 3′–5′ exoribonucleolytic activity and potentially regulates the mammalian noncoding transcriptome¹. Here we generated a mouse model in which the essential subunit Exosc3 of the RNA exosome complex can be conditionally deleted. Exosc3-deficient B cells lack the ability to undergo normal levels of class switch recombination and somatic hypermutation, two mutagenic DNA processes used to generate antibody diversity via the B-cell mutator protein activation-induced cytidine deaminase (AID)^2,3. The transcriptome of Exosc3-deficient B cells has revealed the presence of many novel RNA exosome substrate ncRNAs. RNA exosome substrate RNAs include xTSS-RNAs, transcription start site (TSS)-associated antisense transcripts that can exceed 500 base pairs in length and are transcribed divergently from cognate coding gene transcripts. xTSS-RNAs are most strongly expressed at genes that accumulate AID-mediated somatic mutations and/or are frequent translocation partners of DNA double-strand breaks generated at Igh in B cells^4,5. Strikingly, translocations near TSSs or within gene bodies occur over regions of RNA exosome substrate ncRNA expression. These RNA exosome-regulated, antisense-transcribed regions of the B-cell genome recruit AID and accumulate single-strand DNA structures containing RNA–DNA hybrids. We propose that RNA exosome regulation of ncRNA recruits AID to single-strand DNA-forming sites of antisense and divergent transcription in the B-cell genome, thereby creating a link between ncRNA transcription and overall maintenance of B-cell genomic integrity.

AID mutates single-strand DNA (ssDNA) substrates that form during transcription across the B-cell genome. Current DNA targeting models propose that AID binds paused/stalled RNA polymerase II complexes (RNA Pol II) to access target DNA⁶. In turn, RNA Pol II associates with the pausing/stalling cofactors Spt5 and RNA exosome, both of which stimulate AID function in B cells^7–9. Since RNA exosome is a functional component of the stalled RNA Pol II^10,11 targeting platform of AID, we evaluated RNA exosome’s role in regulating AID activity genome-wide. Accordingly, we developed a mouse model containing a conditional inversion (COIN)¹² allele of Exosc3, allowing conditional ablation of RNA exosome function using tissue-specific or inducible Cre recombinase alleles (Fig. 1a). Cre-mediated ablation of Exosc3 with this allele leads to concomitant green fluorescent protein (GFP)reporter induction from the Exosc3 locus (details in Methods and Extended Data Fig. 1). B cells were generated from Exosc3^COIN/+ and Exosc3^COIN/COIN mice on the 4-hydroxytamoxifen (4-OHT)-inducible ROSA26^CreERt2/+ background. 4-OHT treatment of these cells produced robust Exosc3 gene inversion, loss of Exosc3 messenger RNA and protein, and induction of GFP (Fig. 1b, c and Extended Data Fig. 1d–f).

Figure 1. Exosc3-deficient B cells are defective in immunoglobulin diversification.

a, Exosc3^COIN allele and conversion to Exosc3^COINinv. Cre-mediated inversion of lox2372 pair (red triangles) and subsequent deletion via loxP pair (violet triangles). GFP-expressing terminal exon is represented by green arrow. SA, splice acceptor. b, Southern blot of HindIII-digested genomic DNA from 4-OHT-treated (days 2–4), lipopolysaccharide (LPS) plus interleukin (IL)-4 stimulated B cells. Probe specific for Exosc3 exon 3. WT, wild type. c, qRT–PCR time course of Exosc3 mRNA expression in 4-OHT-treated (days 2–4), LPS plus IL-4 stimulated B cells. Indicated Exosc3 genotypes on a ROSA26^CreERt2/+ background. Expression levels normalized to cyclophilin (Ppia) and plotted relative to untreated Exosc3^COIN/+. Three technical replicates, error bars represent standard deviation (s.d.). d, IgG1 CSR efficiency in 4-OHT-treated Exosc3^COIN/+ and Exosc3^COIN/COIN B cells after 72 h of LPS plus IL-4 stimulation. Indicated Exosc3 genotypes on a ROSA26^CreERt2/+ background. Mean values from three biological replicates are indicated. Error bars represent standard error of the mean (s.e.m.). e, Flow cytometric analysis of Peyer’s patch germinal centre B cells. Percentage of B220⁺ PNA^hi germinal centre B cells amongst all B220⁺ cells is indicated. Experiment was replicated three times. f, SHM analysis of Peyer’s patch derived GFP⁺ germinal centre B cells at AID substrate CG base pairs at the JH4 intron. Mean values determined from 197 (Exosc3^COIN/+) and 203 (Exosc3^COIN/COIN) sequence clones are indicated. *P < 0.01 (t-test), **P < 0.01 (proportion test).

We evaluated CSR efficiency in ex vivo cultured B cells upon 4-OHT-mediated ablation of Exosc3. Immunoglobulin (Ig)G1 class switch recombination (CSR) was decreased approximately fourfold in Exosc3^COIN/COIN B cells compared to littermate control Exosc3^COIN/+ B cells (Fig. 1d and Extended Data Fig. 2a) despite comparable AID expression and increased nascent IgSγ1 transcription (Extended Data Figs 1f and 2b, c). To determine RNA exosome involvement in somatic hypermutation (SHM), we generated Exosc3^COIN/+ and Exosc3^COIN/COIN mice expressing Cre recombinase at early (Cd19^Cre) and late stages (Aicda^Cre) of B-cell development (Aicda^Cre allele details in Extended Data Fig. 2d–f). Cd19^Cre-mediated ablation of Exosc3 leads to B-cell developmental arrest preceding the germinal centre reaction (Extended Data Fig. 2h). However, Aicda^Cre-mediated deletion permits robust germinal centre B-cell production in Exosc3^COIN/COIN mice, with a moderate increase in cell number compared to Exosc3^COIN/+ mice (Fig. 1e). The kinetics of GFP induction and maintenance between Exosc3^COIN/+Aicda^Cre/+ and Exosc3^COIN/COINAicda^Cre/+ B cells ex vivo demonstrated little to no visible growth advantage between deleted (GFP⁺) and non-deleted (GFP⁻) cells (Extended Data Figs 3a, b). VPD450 dye dilution assays demonstrated comparable proliferation between Exosc3^COIN/+ Aicda^Cre/+ and Exosc3^COIN/COINAicda^Cre/+ B cells (Extended Data Fig. 3c, d). We determined the inversion efficiency of Exosc3^COIN in sorted Exosc3^COIN/COINgerminal centre B cells to be~70%, compared to nearly complete inversion in Exosc3^COIN/+ (Extended Data Fig. 1g). SHM downstream to the Igh J_H4 exon was evaluated in Exosc3^COIN/+Aicda^Cre/+and Exosc3^COIN/COINAicda^Cre/+germinal centre B cells. Total mutation frequency was reduced in Exosc3^COIN/COIN mice (Extended Data Fig. 2g) and exacerbated at direct AID target dC:dG base pairs (53% of Exosc3^COIN/+, P < 0.01) (Fig. 1f). Importantly, since AID expression precedes Exosc3 deletion in these assays, we expect some SHM and CSR to occur before Exosc3 depletion, thus underrepresenting the complete effect of RNA exosome deletion on SHM and CSR.

Various ncRNA species, particularly those associated with transcription regulation, are substrates of RNA exosome^13–20. To uncover ncRNA substrates of RNA exosome in B cells, we performed whole transcriptome RNA sequencing on Exosc3-deficient cells. We reconstructed the transcriptomes of Exosc3^WT/WT (wild-type) and Exosc3^COIN/COIN B cells and hereafter refer to the Exosc3^COIN/COIN transcriptome as the ‘exotome’ (Fig. 2a). Small nucleolar RNAs (snoRNAs) and small nuclear RNAs (snRNAs), known targets of RNA exosome in Saccharomyces cerevisiae²¹, were upregulated in the exotome (Fig. 2a). The identity, read counts and coordinates of the ncRNAs analysed in Fig. 2a are provided in Supplementary Tables 1–3. Greatly upregulated in Exosc3-deficient B cells were RNA exosome substrate TSS RNAs (xTSS-RNAs) (Fig. 2a). Short ncRNAs arising from TSSs have been shown previously to be RNA exosome substrates in mammalian cells^13,16,22, although their genome-wide distribution and characteristics are not fully understood. xTSS-RNAs are expressed at regions upstream of mRNA-associated TSSs (Fig. 2b and Extended Data Fig. 4a) and either overlap with cognate mRNA TSSs (ungapped xTSS-RNA) or possess distinct TSSs (gapped xTSS-RNA). RNA-sequencing (RNA-seq) reproducibility was statistically strong (ρ = 0.95; Extended Data Fig. 4b).

Figure 2. RNA exosome depletion reveals xTSS-RNAs.

a, Differential expression analysis for various transcript classes. Horizontal bars indicate 95% confidence intervals with Bonferroni adjustment of log₂ fold change between Exosc3^COIN/COIN and Exosc3^WT/WT. Circles indicate mean values. Red bars indicate transcript types expressed significantly higher compared with mRNA (two biological replicates). P values were determined by Wilcoxon rank-sum test. b, Strand-specific distribution of RNA-seq reads upstream and downstream of expressed coding gene TSSs (two biological replicates). c, Boxplot indicating gapped xTSS-RNA length distribution. Median values from two biological replicates are indicated. Whiskers represent 99% of data values. **P < 0.01 (Wilcoxon rank-sum test). d, Gap length distribution between xTSS-RNAs and coding transcript TSSs. y-Axis indicates number of xTSS-RNAs in each gap length group (x-axis). Compiled from two biological replicates. e, Percentage of TSS-RNA expressed at indicated reads per kilobase per million reads (RPKM). Compiled from two biological replicates.

In Exosc3^WT/WT cells, xTSS-RNA average length was ~600 bp, whereas in Exosc3^COIN/COIN cells xTSS-RNAs were slightly longer (Fig. 2c and Extended Data Fig. 4c). Average TSS distance between xTSS-RNA and cognate mRNA was ~150 bp (Fig. 2d). Many genes in Exosc3^WT/WT cells display low expression of xTSS-RNA (Fig. 2e). However, Exosc3 deletion results in a shift towards higher xTSS-RNA expression (Fig. 2e). Strand-specific RNA-seq experiments demonstrated that xTSS-RNA transcription largely occurs antisense to mRNA transcription genome-wide (Fig. 2b). While sense genic transcripts are comparable between Exosc3^WT/WT and Exosc3^COIN/COIN, TSS antisense transcripts in Exosc3^COIN/COIN are approximately fourfold higher (Fig. 2b). Gapped xTSS-RNA expression correlated poorly with cognate mRNA expression genome-wide (ρ = 0.11; Extended Data Fig. 4d). Furthermore, xTSS-RNA is not uniformly expressed across the B-cell genome. Actively transcribed genes devoid of xTSS-RNA expression include β-actin, Il2rg and Ung (Extended Data Fig. 4e–g). Collectively, we have identified divergently transcribed anti-sense xTSS-RNAs expressed from a subset of transcribed genes within B cells.

AID introduces mutations within Igh switch sequences during CSR. Upstream of the inducible CSR-specific Igg1 (also known as Ighg1) germline transcript we observed strong accumulation of xTSS-RNA in Exosc3-deficient B cells (Fig. 3a). Expression of Igg1 xTSS-RNA was confirmed through quantitative polymerase chain reaction with reverse transcription (qRT–PCR) and northern blotting (Extended Data Fig. 5a, b). Furthermore, we observed robust xTSS-RNA expression at the AID target genes Myc, Pax5, Cd83, Pim1 and Cd79b (Fig.3b and Extended Data Fig. 5c–f). xTSS-RNA transcription at these genes was largely antisense (Extended Data Fig. 6a).

Figure 3. xTSS-RNA expression marks AID-dependent translocation hotspots in the B-cell genome.

a, Antisense Igg1 xTSS-RNA. Strand-specific RNA-seq reads 2 kb upstream and downstream of Igg1 germline transcript TSS. Compiled from two biological replicates. Chr, chromosome. b, Profile of RNA-seq reads at Pim1 locus (8.6 kb window). Red arrow indicates xTSS-RNA. Four biological replicates. c, Venn diagram of genes with xTSS-RNAs, genes undergoing recurrent AID-dependent chromosomal translocations⁴ and somatically hypermutated genes²³. xTSS-RNA group compiled from four biological replicates. P values determined by Fisher’s exact test. d, Enrichment of gapped xTSS-RNA expression amongst translocation hotspots in Exosc3-deficient B cells. Translocation hotspot gene set comprises 40 genes (identified in c) reported to undergo recurrent AID-mediated translocations displaying higher xTSS-RNA expression. ‘Other xTSS-RNA genes’ set comprises 1,694 genes expressing both xTSS-RNA and cognate mRNA, but not reported as recurrent translocation hotspots. Median values from two biological replicates are indicated. **P < 0.01 (Wilcoxon rank-sum test). e, Myc translocation breakpoints at sites of xTSS-RNA and genic antisense transcription. Mouse B-cell translocation frequency⁴ and antisense transcription are shown on the positive and negative y-axes, respectively. Compiled from two biological replicates. f, Correlation between breakpoints and antisense expression (2 kb upstream of TSS to transcription end site) at translocation hotspots (two biological replicates). Error bars indicate 95% confidence interval and blue line represents robust fit of expected values. Pearson correlation (PCC) is indicated. g, h, Probability of translocation breakpoints with respect to antisense (g) or sense (h) transcription levels. Pearson correlation is indicated.

Recent studies using translocation capture sequencing (TC-Seq) or high-throughput genome-wide translocation sequencing (HTGTS) have identified target genes undergoing recurrent translocations to Igh due to AID-generated DNA breaks in B cells^4,5. These analyses have revealed frequent translocations occurring near the TSSs of actively transcribed genes^4,5. We queried these data sets to determine whether recurrent translocation partners of Igh express xTSS-RNAs. Our analysis revealed a positive statistical correlation between genes expressing xTSS-RNA and recurrent AID-dependent translocation (P < 0.0001; Fig. 3c and Extended Data Fig. 6b). Specifically, 40 genes were identified through this analysis (Fig. 3c and Supplementary Tables 4, 5) and, collectively, xTSS-RNA expression was fourfold higher in Exosc3^COIN/COIN B cells (P < 0.01; Extended Data Fig. 6c). Even amongst all other xTSS-RNA-expressing genes, this group of 40 genes displayed higher xTSS-RNA expression (P < 0.01; Fig. 3d and Extended Data Fig. 6d). In contrast, we observed no difference in collective mRNA expression for these 40 genes between Exosc3^WT/WT and Exosc3^COIN/COIN B cells (Extended Data Fig. 6c). Overlapping a list of genes previously shown to undergo AID-mediated hypermutation in mouse B cells²³ with these 40 xTSS-RNA expressing translocation hotspots revealed 5 genes, consisting of Myc, Pax5, Cd79b, Cd83 and Pim1 (Fig. 3c). Mutation of these genes has been observed in diffuse large B-cell lymphoma patients^24,25. Of the 88 translocation hotspots, 74 contain either TSS-RNA or antisense transcription (Extended Data Fig. 7a). Statistical bootstrapping analysis of 10,000 control sets of 88 genes with similar transcription levels as the translocation hotspot gene setwould randomly select only 15 xTSS-RNA expressing genes, thus validating that the observed group of 40 xTSS-RNA-containing genes at translocation hotspots is not solely determined by the level of cognate mRNA transcription (P < 0.01; Extended Data Fig. 6e). We propose that genes undergoing divergent transcription resulting in RNA exosome recruitment, as evidenced through the presence of xTSS-RNAs, are preferentially targeted by AID.

Many translocations also occur within gene bodies and cannot be readily explained by TSS-proximal transcription. However, RNA exosome can also regulate the expression of antisense RNA (asRNA) initiating within gene bodies. Strikingly, a considerable number of Myc translocation junctions precisely map to a region within intron 1 that expresses RNA exosome substrate asRNA (Fig. 3e; additional examples in Extended Data Fig. 7b and Supplementary Fig. 1). Many of the 48 translocation hotspot genes that do not express xTSS-RNAs (Fig. 3c) do express RNA exosome substrate asRNA (Supplementary Fig. 1). Moreover, breakpoints within these translocation hotspots strongly correlate with the presence of RNA exosome substrate asRNA (ρ = 0.79; Fig. 3f). Additionally, Cd83 translocation breakpoints mapping to regulatory regions contain RNA exosome substrate asRNA (Extended Data Fig. 7c). When analysing all asRNAs in Exosc3-deficient B cells (Extended Data Fig. 8a), we note a strong positive correlation between asRNA expression and the probability of observing translocation breakpoints (r 5 0.89; Fig. 3g), whereas sense transcription correlated poorly (ρ = 0.38; Fig. 3h). Altogether, we provide evidence that AID target sites in the B-cell genome possess RNA exosome substrate asRNA, both TSS-proximal and within gene bodies.

An outstanding question concerns the molecular mechanisms relating AID targeting to genes expressing xTSS-RNAs or asRNAs. Divergent promoters, including those expressing xTSS-RNAs, occupy two pre-initiation complexes positioned divergently and separated by ~150 bp. Divergent transcription enhances localized DNA melting surrounding the two TSSs as polymerases initiate transcription, thus generating ssDNAstructures²⁶. Antisense transcripts from such promoters undergo early termination leading to RNA exosome recruitment²⁷. When stalled antisense transcripts are not efficiently removed, stabilization of transcription-coupled R-loops can prolong ssDNA structures. AID requires ssDNA substrates for recruitment and subsequent DNA deamination³ which is further enhanced by pausing and/or stalling of RNA Pol II⁶. Using chromatin immunoprecipitation (ChIP) assays we find that Exosc3 promotes AID occupancy at target genes Pim1, Pax5, Myc, Cd79b and Cd83 (Fig. 4a). This observation cannot be explained simply through differences in gene expression, as H3K4me3 abundance is similar between Exosc3^WT/WT and Exosc3^COIN/COIN B cells, indicating comparable transcription initiation (Fig. 4b). Consistent with these observations, xTSS-RNA expression is enriched at AID- and Spt5-occupied genes in B cells (Extended Data Fig. 8b). Similarly, xTSS-RNA is also enriched at AID target genes identified by replication protein A sequencing (RPA-seq)²⁸ (Extended Data Fig. 8c), a marker of ssDNA. H3S10ph is a chromatin mark associated with ssDNA-containing R-loopstructures²⁹. We observe that AID-targeted divergent promoters accumulate H3S10ph and Exosc3, unlike robustly transcribed non-divergent promoters (Fig. 4c and Extended Data Fig. 9a–e). H3S10ph accumulation is further enhanced at divergent promoters Pim1, Pax5, Cd79b and Cd83 upon loss of Exosc3 (Fig.4d). ssDNA accumulation was also evaluated using the RNA–DNA hybrid-specific S9.6 antibody³⁰. We find that AID-targeted divergent promoters (Pim1, Cd79b, Cd83) accumulate RNA–DNA hybrids more strongly than transcribed non-divergent promoters (March2, Cmas, Atp13a2) (Fig. 4e). Finally, deletion of xTSS-RNA regions corresponding to Pim1 or Cd83 (Extended Data Fig. 9f–i) impairs AID recruitment to these genes (Fig. 4f,g) and reduces AID-mediated hypermutation within the first kilobase pair of Cd83 (Extended Data Fig. 9j).

Figure 4. AID recruitment to RNA–DNA hybrid-forming divergently transcribed genes.

a, Exosc3 promotes AID targeting to divergent promoters. ChIP was performed using anti-AID or control IgG. qPCR was performed using TSS upstream-specific primers. PCR amplification for the non-divergent genes Cmas, March2, Atp13a2 and Fam107b was below the detection limit in anti-AID ChIP. Three pairs of mice of each genotype were used. Mean values from three technical replicates are indicated. Error bars represent s.d. *P < 0.05, **P < 0.01 (log₂ transformation of Z-test). b, H3K4me3 is unaltered in Exosc3-deficient B cells. ChIP was performed using anti-H3K4me3 (Millipore) or control IgG. qPCR and data analysis are as described in a. Two pairs of mice of each genotype were used. NS, not significant (t-test). c, d, H3S10ph accumulation at divergent promoters. ChIP was performed using anti-H3S10ph (Millipore) or control IgG. qPCR was performed as described in a. H3S10ph enrichment relative to input in Exosc3^WT/WT (c) or fold change of Exosc3^COIN/COIN is indicated (d). Three pairs of mice of each genotype were used. Mean values from three technical replicates are indicated. c, **P < 0.01 (t-test) for H3S10ph ChIP between divergent and non-divergent promoter sets. d, *P < 0.05, **P < 0.01 (log₂ transformation of Z-test). e, RNA–DNA hybrid accumulation at divergently transcribed genes in Exosc3-deficient B cells. qPCR was performed as described in a. Two pairs of mice of each genotype were used. Mean values from three technical replicates are shown. Error bars represent s.d. **P < 0.01 (t-test) for Exosc3^COIN/COIN DNA:RNA immunoprecipitation (DRIP) between divergent and non-divergent promoter sets. f, g, Deletion of xTSS-RNA-expressing region impairs AID targeting to divergently transcribed genes. Anti-AID ChIP was performed on parental or clonal CH12F3 B-cell lymphoma lines containing CRISPR-mediated deletions of Cd83 (f) or Pim1 (g) xTSS-RNA-expressing regions. qPCR was performed using exon 1 specific primers. Mean values from three technical replicates are indicated. Error bars represent s.d. *P < 0.05, **P < 0.01 (t-test). Indicated genotypes (except f, g) are on a ROSA26^CreERt2/1 background and B cells were treated with 4-OHT and stimulated with LPS plus IL-4.

On the basis of our findings, we propose that divergent TSSs generating RNA exosome substrates have a key role in recruiting AID and generating ssDNA structures across the B-cell genome (Extended Data Fig. 10). In conjunction with ssDNA formation, increased RNA Pol II stalling at divergent promoters may further facilitate AID recruitment. Similarly, asRNAs generated within gene bodies can also create ssDNA structures, serving as AID substrates and leading to chromosomal translocation (summarized in Extended Data Fig. 10). Our study provides evidence that in addition to RNA Pol II stalling and ssDNA generation, antisense transcription is important for AID targeting throughout the B-cell genome.

METHODS

Generation of the Exosc3^COIN allele

Bacterial homologous recombination methodologies³¹ were used to modify a bacterial artificial chromosome (BAC) containing the mouse Exosc3 locus (clone bMQ386a13). Two sequential BAC modifications were performed. First, lox2372 and loxP sites were inserted between Exosc3 exons 1 and 2. Subsequently, the COIN module (antisense to Exosc3), a second set of lox2372 and loxP sites, and an FRT-flanked neo selection cassette were inserted between Exosc3 exons 3 and 4. The COIN module is comprised of a 3′ splice acceptor sequence, followed by a T2A-GFP open reading frame, followed by a polyadenylation sequence. Correctly modified Exosc3^COINneo BAC clones were identified by PCR screening across all four recombination junctions and verified by restriction digestion followed by pulse field electrophoresis. The entire 6.9 kb region between the upstream and downstream homology arms of the Exosc3^COINneoBAC clone used for embryonic stem (ES)-cell targeting was confirmed by sequencing. The BAC-based Exosc3^COINneo targeting vector was electroporated into ROSA26^CreERt2/+, 129S6/SvEvTac3×C57BL/6Tac hybrid ES cells and correctly targeted clones were identified by a loss of allele assay³². Exosc3^COINneo/+ chimaeric mice were generated by ES-cell microinjection of blastocysts and crossed with Tg(ACTB:FLPe) mice to excise the neo cassette and produce germline transmission of the Exosc3^COIN allele. The FLPe transgene was subsequently bred out for the production of all Exosc3^COIN experimental cohorts. All mouse experiments were performed in accordance with approved Columbia University Institutional Animal Care and Use Committee protocols.

Cell culture and CSR

Splenic B cells from sex-matched littermate mice were prepared using CD43 microbead (Miltenyi Biotec) negative selection and cultured in RPMI 1640 containing 15% FBS. Ex vivo CSR cultures using the ROSA26^CreERt2 allele were cultured for 16 h with 100 nM4-hydroxytamoxifen (Sigma) and 20 μg ml⁻¹ LPS (Sigma) followed by the addition of IL-4 (R&D Systems) at 20 ng ml⁻¹, and cultured for an additional 72 h. CSR culture conditions using the Aicda^Cre allele were identical with the exception that 4-hydroxytamoxifen was not used. Cells were stained using fluorescent antibodies against B220 and IgG1 (BD Biosciences). Data were acquired on a FACSAria cell sorter (BD Biosciences) and analysed using FloJo software (Tree Star). Antibodies used for western blot purposes: Exosc3 (Santa Cruz Biotechnology), actin (Abcam) and goat anti-rabbit IgG HRP (Jackson ImmunoResearch). All experiments involving mice were performed with known genotypes and therefore performed unblinded. Biological replicates involve B cells isolated from separate mice.

RNA preparation and qPCR

Total RNA was isolated from cells using Trizol reagent (Life Technologies). RNA was resuspended in water and quantified using a Nanovue Plus spectrophotometer (GE Healthcare Life Sciences). RNA samples were treated with DNase I (Turbo DNA-free kit, Life Technologies), eluted in water and re-quantified. 1.5 μg of RNA were then converted to cDNA using random hexamers and the Superscript III First-Strand Synthesis System for RT–PCR (Life Technologies). Exosc3 and Aicda mRNA measurements were performed using the TaqMan Gene Expression Assay (Applied Biosystems). TaqMan assay ID numbers for Exosc3 and Aicda were Mm01345308_m1 and Mm00507774_m1, respectively. All other quantitative RT–PCR experiments were performed using SYBR Green Master Mix (Roche Applied Science). Primer pairs for the quantification of associated xTSS-RNAs were as follows: Pim1, 5′-CACATGCACGTGGAAATACCA-3′ and 5′-CATCCATAAAAGTTATGGAGTC-3′; Pax5, 5′-CTGCTTTTTCAGGTCTAGCTC-3′ and 5′-CCCATTCAAAAGCTCATTAAG-3′; Il4ra, 5′-GGCTGTTGCTCATTTTCCCAA-3′ and 5′-TGTGGGCGAGAGAACAACTTC-3′; Myc, 5′-AGCGCAGCATGAATTAACTGC-3′ and 5′-GTATACGTGGCAGTGAGTTG-3′; Igg1, 5′-GTATCTTGTTGGTGCTATCTCA-3′ and 5′-TGGGATCTGCTACACAGGTTT-3′. Expression levels for individual transcripts were normalized against β-actin and/or cyclophilin with similar results. Fold change in transcript levels were calculated as fold change = 2^{(CtWT,GOI−CtCOIN/COIN,GOI)}/2^{(CtWT,actin−CtCOIN/COIN,actin)}. Ct, cycle threshold. GOI, gene of interest.

Northern blotting

Total RNA (7.5 μg) isolated from splenic B cells was electrophoresed in denaturing conditions on a 1% agarose-formaldehyde gel at 5.5 V cm⁻¹. The gel was rinsed several times in deionized water followed by 20× SSC and then transferred overnight by capillary transfer onto an Amersham Hybond-XL membrane. The RNA was crosslinked to the membrane using a Stratagene UV crosslinker and then stained with methylene blue in 0.3 M sodium acetate to ascertain transfer efficiency. Subsequently the membrane was prehybridized for 5 h before being hybridized with a cDNA encoding the gene of interest (approximate size of probe: 400 bp), radiolabelled through the random primed labelling technique (High Prime, Roche Applied Science) and purified twice over Probequant G50 microcolumns (GE Healthcare) before addition to the hybridization reaction. The membrane was hybridized over-night at 42°C in hybridization solution before being washed twice in 0.1% SDS/2× SSC at room temperature followed by two washes in 0.1% SDS/2× SSC at 65°C, all washes for15 min. Membranes were subsequently exposed to film for varying amounts of time.

RNA-seq analysis

rRNA-depleted total RNA was prepared using the Ribo-Zero rRNA removal kit (Epicentre). Libraries were prepared with Illumina TruSeq and TruSeq Stranded total RNA sample prep kits, and then sequenced with 50–60 million of 2× 100 bp paired raw passing filter reads on an Illumina HiSeq 2000 V3 instrument at the Columbia Genome Center. To construct the transcriptome of Exosc3^WT/WT and Exosc3^COIN/COIN B cells, we first mapped all reads of total RNAs to the mouse reference genome (mm9) with TopHat (v. 1.3.2)³³. Cufflinks (v. 1.2.1) was subsequently applied to assemble the whole transcriptome and to identify all possible transcripts³⁴. To obtain short RNAs (potential exosome targets), we set the overlap radius as 1 and merged repeated samples. All resulting transcripts are further annotated under the supervision of a comprehensive collection of RNA databases including mRNA, snoRNA, long intergenic noncoding RNA (lincRNA), microRNA and other annotated non-coding RNAs (Supplementary Table 1). Specifically, we overlapped all assembled transcripts with known RNAs in the collected database, and annotated the transcripts by their adjacent known RNAs. A transcript is annotated as one isoform of a known RNA if both the TSS and transcription end site (TES) of the transcript are close (less than 200 bp) to those of the known RNA. If comparing with a known RNA, and the TSS of a transcript is shifted less than 200 bp, this transcript is annotated as the mixture of an upstream transcript (ungapped xTSS-RNA) and a known RNA. Similarly, if the TES is shifted more than 200 bp, the corresponding transcript is annotated as a read through. Transcripts that have no overlap with any known RNA located within 2,000 bp upstream of a known RNA are annotated as gapped upstream transcripts (gapped xTSS-RNA). Conjoint transcripts were annotated on the basis of containing sequences from multiple known RNAs. Biological replicates indicate that each RNA-seq data set was generated from B cells isolated from separate mice. B-cell translocation hotspots were defined based on supplementary table of Klein et al.⁴. AID and Spt5 binding loci are based on ChIP-seq data from Pavri et al.⁷.

Statistical analysis

To test the significance of the difference of two vectors, a two-sided nonparametric Wilcoxon rank-sum test was applied to calculate the P values. To test the difference between two proportions, the following equation was used:

$$Z=\frac{p_1-p_2}{\sqrt{P(1-P)\left(\frac{1}{n_1}+\frac{1}{n_2}\right)}}$$

where p₁ and p₂ are two proportions, P is the expected value, and n₁ and n₂ are the population size. P values were then generated by normal distribution. We applied pipeline of DEseq for the normalization of library size, and gene differential expression analysis³⁵. Multiple comparison analysis was performed in MATLAB R2011a. One-way analysis of variance (ANOVA) was followed by multiple comparison procedure with critical values from the t distribution, after a Bonferroni adjustment.

SHM

Peyer’s patches were excised from 2.5–3.5-month-old paired littermates and gently dissociated by passage through a 70 μm cell strainer. Germinal centre B cells were stained with anti-B220 (BD Biosciences) and peanut agglutinin (Vector Laboratories). DAPI stain was added just before cell sorting to exclude dead cells. Cells were directly sorted into lysis buffer containing proteinase K (Viagen) a using FACSAria cell sorter (BD Biosciences) and incubated at 55 °C overnight. Fifteen micrograms of GlycoBlue (Life Technologies) were added to the lysates and genomic DNA was purified via ethanol precipitation. JH4 intron was amplified by PCR using LA Taq (Takara) and a primer pair (J558FR3Fw: 5′-GCCTGACATCTGAGGACTCTGC-3′ and JH4intronRv: 5′-CCTCTCCAGTTTCGGCTGAATCC-3′) that requires a VDJ rearrangement of the Igh locus for amplification to occur³⁶. JH4 amplicons were cloned into the pCR2.1-TOPO vector (Life Technologies) and sequenced using M13 primers. Mutation analysis at Cd83 in CH12F3 cells was performed on a 488 bp sequence between PCR primers GCCTCCAGCTCCTGTTTCTA and TGTTGCTTTCACTGCAGCTCTC.

Analysis of TC-Seq

To capture genome wide the translocation breakpoints of B cells, we downloaded the TC-Seq data sets⁴ from the Sequence Read Archive (http://www.ncbi.nlm.nih.gov/sra) under accession number SRA039959, and followed a similar computational workflow as described by Oliveira et al.³⁷. All alignments are performed by BWA³⁸ with default parameters, and the counting of reads is performed in bedtools³⁹.

Translocation breakpoint probability modelling

To predict the probability of observing breakpoints in a given genomic region, we assume the number of TC-Seq reads in this region follows a negative binomial model with parameters r and p. We applied a maximal likelihood method to estimate the parameters, as well as the confidence interval of the parameters. For a given genomic region, the probability of occurrence of translocation breakpoints is then defined as the probability of harbouring more than x TC-Seq reads. In this manuscript, two types of regions are separately considered. In one type of region, each base pair from 2 kb upstream of the TSS to the TES of translocation hotspot genes is binned into ~10 tiers according to their expression level of antisense transcripts. Approximately 4× 10⁶ genomic positions were binned according to antisense RNA expression level. Negative binomial distribution fitted in each region to estimate breakpoint probability. In the other type, genomic regions of all collected genes that do not harbour known antisense RNAs (28,947 genes) are binned into 46 tiers according to the expression level of their antisense transcripts to estimate the parameters of the negative binomial model.

ChIP

Crosslinking was performed on cultured cells using formaldehyde. Sonication was performed on ice using a Branson Sonifier 250 apparatus for 25 cycles, each cycle comprising 20 s of sonication at duty cycle 30% followed by a 2 min rest period. Lysates were pre-cleared for 1 h. Immunoprecipitation of lysates was performed overnight at 4 °C using indicated antibodies. SepharoseA/G beads were added for 90 min with continued rotation. Subsequently the beads were washed by the standard series of washes (low salt, high salt, LiCl, and TE) and ChIP products were eluted followed by RNaseA treatment overnight at 60 °C and proteinase K treatment for 2 h at 55 °C. ChIP DNA was recovered using ethanol precipitation. Primers used for ChIP quantitative PCR were as follows. Myc, CGGTTGATCACCCTCTATCACTC and GCTCCACACAATACGCCATGTAC; Pim1, CCCAGGATCTAGCCCACATAACATC and AGCGTAGCAAGTTGTGAGAAATGG; Pax5, CTGCTAGGATGGTTCTGCTTGG and CAACTCAATTGCAACCTCCATAGGTC; Cd79b, TGCTGATTGAGAAGGTTGGTGTG and GGAAGGGGTTGCTCCTGAATC; Cd83, AGATCTCCCTTGCTCAAACAACG and GACCTGCTACTCTCCAGATTTTGTG; Cmas, GGAAAACGGAAAGAGGCTGGAG and TGAGCTCAGAGGAGCCTCTAG; Atp13a2, CAGCCTGTCCTTTTCCGTCTATC and AGCTCGCTGAGATCTTGATGC; March2, GCAGCAAGTCTACAGCCAGAG and GCCTCTGAGTATCATCTGCCAATC; Fam107b, GACACCTTCCATTAGACAGGTGAC and AGATGAGAGCTCTGGATCCTTGG. Technical replicates of ChIP were performed from the same cell type.

DRIP

DRIP was performed on cultured B cells following a previously described protocol³⁰. Briefly, cells were spun down and digested overnight in TE with proteinase K, followed by phenol-chloroform extraction and ethanol precipitation. Genomic DNA was digested using BsoBI, NheI, NcoI and StuI either with or without RNase H (NEB). Purified DNA was then immunoprecipitated using S9.6 antibody (gift from F. Chédin). Technical replicates of DRIP were performed from the same cell type.

CRISPR/Cas9-mediated targeted deletions

Guide RNA sequences were designed using an online tool (http://tools.genome-engineering.org). Guide RNA-encoding oligonucleotides (Cd83, AGTGCCCAACACTACCTAAT and TTCCGAAGCCTCAGGGCGCG; Pim1, ATCAGACACATTCCGAGAAG and CTCTGTGTTTCCCGGAGATT) were cloned into the BbsI site of pSpCas9(BB)-2A-GFP (pX458 Addgene) as described⁴⁰. Guide RNA/Cas9 expression vectors were electroporated into CH12F3 cells using Amaxa Nucleofector (Lonza). Cells were cloned using limiting dilution 3 days after electroporation. Individual clones were screened for homozygous deletion of xTSS-RNA-encoding regions using PCR. Screening primers for Cd83 xTSS-RNA deletion were CCATGCTACAATGCACAGACCTAC and CAGCCTAGAAACAGGAGCTGGAG. Screening primers for Pim1 xTSS-RNA deletion were CCAGGGATCAAACCTAGGATTTTC and CAGAAGACGCCCTATTTGCATAAGG. AID ChIP primers were as follows: Pim1, CTCGCTCCGCCGCCGCTGCTG and CGCAGGTGGGCCAGGGAGTTGAT; Cd83, GCCTCCAGCTCCTGTTTCTA and TCGGAGCAAGCCACCGTCAC.

Extended Data

Extended Data Figure 1. Exosc3 gene targeting and functional validation of the Exosc3^COIN allele.

a, Schematic of the Exosc3^COINneo BAC targeting vector. Blue shaded boxes indicate Exosc3 exons 1–4. Lox sites are represented by triangles. The GFP-expressing gene trapping module is represented by green arrows. Upstream, downstream and internal homology arms are 41.3, 81.5 and 2.4 kb, respectively. b, Confirmation of Exosc3^COINneo/+ targeted embryonic stem (ES)-cell clone AF8. The loss-of-allele (LOA) assay³² was used to screen ES-cell clones for wild-type allele copy number at defined locations within Exosc3 introns 1 and 3 that have been modified to allow for distinction between wild-type and COINneo alleles by TaqMan-based qPCR. A probe for a non-targeted locus, Wnt2b, served as an internal qPCR standard for both copy number and total input DNA. Data represent mean values from six technical replicates. Error bars represent s.d. Ct, cycle threshold. c, HindIII restriction map of the wild-type (WT), COIN and COINinv alleles of Exosc3. The black shaded box indicates the location of the probe used for Southern blotting in Fig 1b. d, Flow cytometric analysis of GFP expression in naive or 4-OHT-treated, LPS plus IL-4-stimulated B-cell cultures. Indicated Exosc3 genotypes are on a ROSA26^CreERt2/+ background. One pair of littermate mice was used. Three biological replicates were performed. e, Profile of RNA-seq mapped reads at the Exosc3 locus from 4-OHT-treated, LPS plus IL-4-stimulated B-cell cultures. Indicated Exosc3 genotypes are on a ROSA26^CreERt2/+ background. Four biological replicates were performed. f, Immunoblot analysis of Exosc3 and AID protein expression in whole cell extracts from 4-OHT-treated, LPS plus IL-4 stimulated B-cell cultures. Actin was used as a loading control. Wild type, Exosc3^WT/WT ROSA26^CreERt2/+; C/C, Exosc3^COIN/COIN ROSA26^CreERt2/+. One pair of littermate mice was used. Two technical replicates were performed. g, Exosc3^COIN:Exosc3^COINinv ratio in germinal centre B cells determined by qPCR copy number analysis (three technical replicates, error bars represent s.d.).

Extended Data Figure 2. Exosc3-deficient B cells are impaired in CSR and SHM.

a, Representative flow cytometric analysis for surface IgG1 on purified B cells treated with 4-OHT, and stimulated with LPS plus IL-4. Numbers indicate the percentage of GFP⁺ B220⁺ B cells having isotype switched to IgG1. One pair of littermate mice was used. Three biological replicates were performed. b, Quantitative RT–PCR time-course analysis of Aicda mRNA expression in naive (day 0) or 4-OHT-treated (days 2–4), LPS plus IL-4 stimulated B-cell cultures. Indicated Exosc3 genotypes are on a ROSA26^CreERt2/+ background. Expression levels are normalized to cyclophilin (Ppia) and plotted relative to naive Exosc3^COIN/+. Six littermate pairs of each genotype were used. Data represent mean values from three technical replicates. Error bars represent s.d. **P < 0.01 (t-test). c, Quantitative RT–PCR analysis of Ighg1 switch region intron expression. Primers were designed to amplify a region of the Ighg1 GLT intron upstream of the Ighg1 switch region core repeat, but downstream of the Ighg1 non-coding I exon. Two independent pairs of littermate mice of each genotype were used to obtain total RNA from B-cell cultures treated with 4-OHT and stimulated with LPS plus IL-4. Indicated genotypes are on a ROSA26^CreERt2/+ background. Data represent mean values from three technical replicates. Error bars represent s.d. Two biological replicates were performed. d, Schematic of the targeted Aicda^Cre allele. An open reading frame comprising a nuclear localization signal fused to Cre recombinase was used to disrupt the ATG start codon in exon 1 of Aicda. Exons are represented as numbered boxes. e, Specific induction of Aicda^Cre activity upon LPS plus IL-4 stimulation of B cells. Flow cytometric analysis of Aicda^Cre activity (as determined by GFP expression) in B220⁺ and B220⁻ naive splenocyte populations (top panel). Aicda^Cre induction in LPS plus IL-4 stimulated B-cell cultures (bottom panel). One pair of littermate mice was used. f, Aicda^Cre is a functional null allele. CSR to IgG1 isotype is abrogated in Aicda^Cre/Cre homozygous B cells stimulated with LPS plus IL-4. Numbers above gate indicate the percentage of GFP⁺ B cells having isotype switched to IgG1. One pair of littermate mice was used. g, SHM analysis of Peyer’s patch derived GFP⁺ germinal centre B cells. Mutation frequencies were determined by sequencing a 645 bp intronic region downstream of the JH4 gene segment of the immunoglobulin heavy chain (IgH) locus. Two littermate pairs of each genotype were used. Two biological replicates were performed. Mutation frequencies represent mean values. P values were determined by proportion test. h, Flow cytometric analysis of Peyer’s patch derived germinal centre B cells from Exosc3^COIN/+ and Exosc3^COIN/COIN mice on a Cd19^Cre/1 background were identified as B220⁺ PNA^hi populations. The percentage of germinal centre B cells amongst all B220⁺ cells is indicated. One pair of littermate mice was used. Three biological replicates were performed.

Extended Data Figure 3. Proliferation analysis of Exosc3-deficient B cells.

a, FACS analysis indicating the percentage of GFP-negative (left gate) and GFP-positive (right gate) B cells 3.5 days after LPS stimulation. One pair of littermate mice was used. Two biological replicates were performed. b, Kinetic analysis of GFP-positive B-cell accumulation at indicated time points post-LPS stimulation. Indicated Exosc3 genotypes are on a Aicda^Cre/+ background. One pair of littermate mice was used. Two biological replicates were performed. c, Proliferation analysis determined by VPD450 dilution at 1.5 and 2.5 days post-LPS stimulation. One pair of littermate mice was used. d, Quantitative RT–PCR analysis of Exosc3, Aicda and GFP mRNA expression in GFP⁺ cells at 3.5 days post-LPS stimulation. Expression levels are normalized to β-actin and plotted relative to Exosc3^COIN/+. One pair of littermate mice was used. Data represent mean values from three technical replicates. Error bars represent s.d.

Extended Data Figure 4. Transcriptome analysis of Exosc3-deficient B cells.

a, Genome-wide expression level analysis upstream and downstream of TSS region for expressed protein coding genes. Coding genes with FPKM >1 were determined to be expressed. Analysis was restricted to coding genes that do not have any known genes within a 4 kb upstream boundary. Indicated genotypes are on a ROSA26^CreERt2/+ background. One sex-matched littermate pair was used. Two biological replicates were performed. b, Replicate analysis of genome-wide studies. Plots indicate the expression levels of individual genes in Exosc3^WT/WT and Exosc3^COIN/COIN B cells treated with 4-OHT and stimulated with LPS plus IL-4 from two separate littermate pairs. B cells were purified, cultured and FACS sorted, and RNA was purified and sequenced by RNA-seq all independently between the two experiments. Indicated genotypes are on a ROSA26^CreERt2/+ background. Pearson correlation is indicated. c, The distribution of observed lengths for all gapped xTSS-RNAs in Exosc3-deficient B cells. Data were compiled from two biological replicates. d, Scatter plot indicating weak correlation between expression of downstream coding transcript and upstream gapped xTSS-RNA at divergently transcribed loci in Exosc3-deficient B cells. Pearson correlation is indicated. e–g, Profile of RNA-seq mapped reads at the β-actin locus (e) (Actb; 7.6 kb window), Il2rg locus (f) (5.4 kb window) and Ung locus (g) (12 kb window). Indicated genotypes are on a ROSA26^CreERt2/+ background and B-cell cultures were treated with 4-OHT and stimulated with LPS plus IL-4. Four biological replicates were performed.

Extended Data Figure 5. xTSS-RNA expression at AID target genes.

a, Quantification of xTSS-RNA expression levels of AID target genes via quantitative RT–PCR from two independent experiments. Indicated genotypes are on a ROSA26^CreERt2/+ background. b, Northern blot analysis of xTSS-RNA expression at Myc and Igsγ1 loci. WT, Exosc3^WT/WT ROSA26^CreERt2/+; C/C, Exosc3^COIN/COIN ROSA26^CreERt2/+. c–e, Profiles of RNA-seq mapped reads at the Pax5 (c) (72 kb window displaying exons 1–6), Cd83 (d) (22 kb window), and Myc (e) (9 kb window) loci. Red arrows highlight the presence of xTSS-RNA. Four biological replicates were performed. f, Quantification of xTSS-RNA expression levels for AID target genes Myc, Igg1, Cd83, Pim1, Pax5 and Cd79b was obtained from RNA-seq RPKM values from two independent experiments. Indicated genotypes are on a ROSA26^CreERt2/+ background. **P < 0.01 (t-test).

Extended Data Figure 6. Translocation hotspots are enriched for xTSS-RNA expression.

a, Strand-specific RNA-seq mapped reads at AID target genes Myc, Cd83, Pim1, Pax5 and Cd79b. Green and red peaks indicate sense and antisense reads, respectively. Red bars represent RefSeq annotation of gene exons. Asterisks indicate the location of TSSs. Arrows indicate the orientation of coding strand transcript. Data were compiled from two biological replicates. b, Boxplot analysis of the level of expression of xTSS-RNAs at various genes reported to undergo recurrent AID-dependent translocations at DNA double-strand breaks generated within the Igh (left panel) or Myc (right panel) loci. Boxplots represent median values compiled from two biological replicates. Whiskers represent 99% of data values. **P < 0.01 (Wilcoxon rank-sum test). c, The list of 40 genes that show an overlap of translocation hotspots and xTSS-RNA expression (from Fig. 3c) was evaluated directly for xTSS-RNA levels (left panel) and mRNA levels (right panel). Statistical analysis was as described in b. **P < 0.01; NS, not significant (Wilcoxon rank-sum test). d, xTSS-RNA expression levels in Exosc3-deficient B cells at non-recurrent and recurrent AID-dependent translocation sites in the B-cell genome. Data were compiled from two biological replicates. **P < 0.01 (Wilcoxon rank-sum test). e, Statistical analysis of the probability of identification of 40 random xTSS-RNA-expressing genes solely based on expression level. Ten-thousand control group genes were randomly selected that were expressed at similar levels as translocation hotspots genes. Specifically, to generate one random control group, we exhausted all translocation hotspots to find genes with similar expression levels (difference of RPKM < 0.5), and randomly picked up one for each hotspot. Ten-thousand gene lists were obtained that contain 88 genes and share the same expression profile with the translocation hotspots list. We then simulate the distribution of genes containing xTSS-RNA by overlapping the random control groups and actual xTSS-RNA gene list. The binomial fitting (red curve) shows that the number of overlapping genes of real translocation hotspots is significantly higher than random controls. **P < 0.01 (binomial distribution).

Extended Data Figure 7. RNA exosome substrate antisense transcripts are expressed within gene bodies and regulatory regions containing AID-induced translocations.

a, Association of genes with TSS-RNA expression, antisense transcripts and AID-induced translocations. The xTSS-RNA and antisense transcripts groups were compiled from four and two biological replicates, respectively. b, Examples of genes with asRNA transcription (Exosc3^WT/WT in blue and Exosc3^COIN/COIN in red) at regions that have been shown to have translocations from the Igh locus (translocations indicated in black). Data were compiled from two biological replicates. c, Translocations present in the upstream regulatory regions of AID target gene Cd83 (top panel) occur over regions of RNA exosome-sensitive antisense transcription. Data were compiled from two biological replicates.

Extended Data Figure 8. Genome-wide analysis of xTSS-RNA expression at genes with AID, Spt5 and RPA occupancy.

a, Scatter plot of antisense (left) and sense RNA transcription (right) in Exosc3^WT/WT and Exosc3^COIN/COIN transcriptomes. Data were compiled from two biological replicates. b, Genome-wide analysis of xTSS-RNA expression at genes that are expressed and possess or lack AID and/or Spt5 occupancy. Values beneath each group represent the number of genes with indicated occupancy. P values were determined by Wilcoxon rank-sum test. c, Overlap of genes with xTSS-RNA transcription (pink), recurrent AID-dependent chromosomal translocations (green) and RPA occupancy in the mouse B-cell genome (brown). The xTSS-RNA group was compiled from four biological replicates. **P < 0.01 (Fisher’s exact test).

Extended Data Figure 9. RNA exosome and AID recruitment to divergently transcribed promoter regions.

a, ChIP was performed using anti-H3S10ph or control IgG. Quantitative PCR and data analysis were performed as described in Fig. 4c. **P < 0.01 (t-test). b–d, Representative plots of highly expressed non-divergent genes used as controls for ChIP experiments in Fig. 4. These genes are Cmas (b), Fam107b (c) and March2 (d). Four biological replicates were performed. e, Exosc3 occupancy at divergent and non-divergent promoters. ChIP was performed using anti-Exosc3 (Genway) or control rabbit IgG. Quantitative PCR was performed using primers specific for sequences upstream of the indicated gene TSS. Data are represented as Exosc3 enrichment relative to input. Data represent mean values from three technical replicates. Error bars represent s.d. **P < 0.01 (t-test). f, g, CRISPR/Cas9-mediated deletion strategy of Cd83 (f) and Pim1 (g) xTSS-RNA-expressing regions in CH12F3 B cells. Locations of CRISPR/Cas9 guide RNAs (red markings), genotyping primers (blue triangles), ChIP primers (green triangles), and numbered exons (blue boxes) are indicated. h, Genotyping of Cd83 and Pim1 xTSS-RNA region-deleted CH12F3 clones. i, Cd83 and Pim1 mRNA expression in xTSS-RNA region-deleted CH12F3 cells. Data represent mean values from three technical replicates. *P < 0.05 (t-test). j, Deletion of Cd83 xTSS-RNA-expressing region impairs SHM. Parental CH12F3 or Cd83 xTSS-RNA region-deleted CH12F3 cells were transduced with lentiviral AID and mutation frequency was determined within a 488 bp region beginning approximately 150 bp downstream of the Cd83 TSS. All mutations were derived from unique clonal amplified sequences. Impairment of Cd83 SHM in Cd83 xTSS-RNA region-deleted cells is disproportionally greater than mRNA expression change observed in i. Number of sequenced clones for parental and Cd83 xTSS-RNA region-deleted CH12F3 was 69 and 102, respectively. Background mutation frequency was determined using uninfected control CH12F3 cells and subtracted from the mutation frequencies indicated. **P < 0.01 (proportion test).

Extended Data Figure 10. A model of RNA exosome recruitment to divergently transcribed promoters or at DNA sequences that promote RNA Pol II stalling.

a, Divergent transcription of mRNA in the sense direction recruits RNA exosome and AID following stalling due to various transcription impediments (G-richness in IgH switch sequences is one example). Transcription stalling leading to RNA exosome recruitment occurs more often on the antisense strand due to formation of short as RNAs²⁷. Similarly, in the body of transcribed genes, stalled RNA Pol II generates asRNA transcripts, leading to RNA exosome and AID recruitment. b, Stalled transcripts either close to the TSS or within the body of genes generate DNA–RNA hybrids. These DNA–RNA hybrids contain RPA-coated ssDNA structures that are targets of AID.