Mutations, kataegis, and translocations in B lymphocytes: towards a mechanistic understanding of AID promiscuous activity

Abstract

As B cells engage in the immune response they express the deaminase AID to initiate the hypermutation and recombination of immunoglobulin genes, which are crucial processes for the efficient recognition and disposal of pathogens, However, AID must be tightly controlled in B cells to minimize off-targeting mutations, which can drive chromosomal translocations and the development of B cell malignancies, such as lymphomas. Recent genomic and biochemical analyses have begun to unravel the crucial question of how AID-mediated deamination is targeted outside immunoglobulin genes. Here, we discuss the transcriptional and topological features that are emerging as key drivers of AID promiscuous activity.

When B cells migrate out of the bone marrow as naïve lymphocytes, they carry substantial alterations at their B cell receptor genes. It is estimated that at the conclusion of RAG-mediated V(D)J recombination in the bone marrow, B cells have diversified their immunoglobulin gene repertoire to the extent that they can recognize an astronomical 5 × 10¹³ different molecules¹. Yet this primary repertoire only represents a fraction of the further diversification that occurs in mature B cells and is mediated by the somatic hypermutation (SHM) of variable (V) domains, which increases the affinity of antibody molecules for the immunogen². In addition to SHM, activated B cells replace their immunoglobulin heavy chain (IgH) constant Cµ domain (IgM isotype) for the constant (C) domain of downstream immunoglobulin isotypes (Cγ, Cα or Cε) which encode the C region for IgG, IgA or IgE respectively. This class switch recombination (CSR) process imparts antibodies with different means to eliminate pathogens and antigens.

Both CSR and SHM are initiated by activation-induced cytidine deaminase (AID; encoded by AICDA), which converts deoxycytidines into deoxyuridines upon recruitment to V and switch (S) recombination sequences^{3, 4}. The uracil base in DNA following deamination engages the activity of base excision repair (BER) and mismatch repair (MMR) pathways, which create nicks and double-strand breaks (DSBs) that initiate SHM and CSR².

AID activity is predominantly restricted to immunoglobulin genes. To a lesser extent however, AID promiscuously targets a subset of transcriptionally active genes, including the proto-oncogenes B cell lymphoma 6 (BCL6) and MYC^5–7. Frequent lesions at these loci can result in mutations or chromosomal translocations, ultimately leading to the dysregulation of the proto-oncogene and B cell tumorigenesis⁷. Because of the importance of AID in both humoral immunity and lymphomagenesis, the molecular basis for the specificity of AID targeting has attracted considerable attention in the past 5 years.

In this review we summarize recent findings that help explain the affinity of AID for immunoglobulin and non-immunoglobulin loci. We first describe deep-sequencing techniques that were developed to assess AID activity across the genome. We follow with a description of the functional and topological features associated with these sites and the potential ways by which they facilitate AID off-targeting. We end with a discussion of some of the mechanisms B cells have evolved to reduce the probability of oncogenic transformation by AID.

AID off-target hypermutation

The process of SHM was initially thought to be restricted to the immunoglobulin loci. However, the BCL6 proto-oncogene was often found mutated at promoter proximal sequences in follicular lymphoma and diffuse large B cell lymphoma (DLBCL)^{8, 9}. A priori, these mutations could be the result of tumor instability and selection. However, BCL6 mutations were also observed in primary lymphocytes from healthy donors. Furthermore, the mutations were mostly restricted to BCL6 intron 1 (which coincides with the SHM targeting area at immunoglobulin loci¹⁰) and displayed the mutation spectrum that is characteristic of immunoglobulin SHM^{5, 6}. Following this initial discovery, additional genes including CD95, CD79A, CD79B, PIM1, MYC, RHOH and PAX5 were also found to be hypermutated in tumors and primary B cells⁷.

Repair pathways

The BER and MMR pathways faithfully repair U-G mismatches downstream of AID. In their absence (for example in Msh2^−/−Ung^−/− mice), replication over unrepaired uracils leads to the accumulation of C to T (and G to A) transition mutations¹¹. Correspondingly, a mutation analysis of selected genes in Msh2^−/−Ung^−/− Peyer’s patches B cells identified additional AID off-targets¹². Intriguingly, this study showed that the protection against AID-mediated mutations conferred by the BER and MMR pathways was locus specific. For instance, whereas Bcl6 mutations were evident in both Msh2^−/−Ung^−/− and wild-type cells, Myc was only mutated when BER and MMR were impaired¹². Why protection against AID-mediated attack by repair pathways varies from gene to gene remains an interesting problem in the field.

Cataloguing the range of AID off-targets

To unravel the true nature of AID’s promiscuous activity, one must first catalogue the full range of AID off-targets. This task is complicated by the fact that DNA occupancy by AID does not predict DNA damage^{13, 14}. Furthermore, early studies only measured SHM by Sanger sequencing at sites of interest. Conversely, the high error rate of high-throughput technologies (~1 miscalled base per 100 nucleotides sequenced¹⁵) has for the most part precluded the use of deep-sequencing to measure SHM genome-wide. A recent method, termed mutational analysis by paired-end deep-sequencing (MutPE-Seq)¹⁶, makes use of long paired-end reads to offset this ‘mutation’ background. However, similar to conventional methods, this technique can only measure mutations at pre-defined sites that are amplified by PCR. An alternative strategy, termed SHM-Seq, involves the microsequencing of cell subclones from SHM-proficient B cell lines, where acquired mutations are present at frequencies similar to single nucleotide polymorphisms (SNPs)¹⁷. In a recent study using this SHM-Seq approach, mutations were measured in DNA associated with trimethylated histone H3 Lys4 (H3K4me3), which is an epigenetic mark that closely overlaps with AID activity. The assay also involved long-term inhibition of MSH2 and UNG and overexpression of AID in the human Ramos–Burkitt lymphoma line, which undergoes trace levels of constitutive hypermutation. Furthermore, the accumulation of mutations was facilitated by enhanced expression of AID under the control of an enhancer-promoter cassette from the immunoglobulin κ-chain gene (Igκ) (first described in REF.¹⁸). This study¹⁷ confirmed that AID mediates the hypermutation of genes implicated in B cell transformation, including BCL6, MYC, BCL7A, MSH6, MIR142 and ID3, and identified several new features of AID off-targeting (discussed below).

Mapping AID off-target DNA breaks and translocations

Long before the discovery of AID, genetic studies revealed chromosomal translocations involving immunoglobulin genes in human B cell tumors⁷. These rearrangements altered the expression of proto-oncogenes due to their juxtaposition to potent immunoglobulin enhancers. Given that DNA is physiologically remodeled at immunoglobulin loci, it was logical to regard DNA breaks intermediate to CSR as substrates for translocations in lymphoma precursors. This hypothesis was tested in BALB/c mice^19–21, in which intraperitoneal injection of pristane oil or transgenic overexpression of IL-6 causes Myc–Igh rearrangements and neoplasia²². The requirement for AID-mediated breaks in these translocations was demonstrated by their substantial reduction in Aicda^−/− or Ung^−/− mice (in the absence of Ung, mutations are not converted into DNA breaks)^19–21.

Still unanswered was whether AID is also responsible for the DSBs at translocating oncogenes. To explore this, restriction sites for the I-SceI endonuclease were knocked-in at Myc and Igh, and artificial DSBs were induced at either locus in B cells undergoing CSR. In this setting, Myc translocation to Igh was detected only in AID-proficient cells²³. Since Myc is only one of many damaged oncogenes in B cell malignancies, two high-throughput technologies (TC-Seq and HTGTS) were developed to systematically identify translocation hotspots involving AID-dependent breaks^{24, 25}. The techniques were applied to ex vivo-activated B cells^{24, 25} and more recently to germinal center B cells from Plasmodium chabaudi-infected mice¹⁶. Altogether, the experiments revealed that AID targets >100 genes, including at oncogenes that are frequently translocated in lymphoma.

TC-Seq and HTGTS revealed translocation hotspots caused by AID. Detecting the actual AID-mediated breaks behind the translocations was more challenging. For instance, attempts to ChIP-Seq non-homologous end joining (NHEJ) factors, such as Nbs1 or γH2AX, did not uncover reproducible or robust hotspots of AID-mediated DNA damage^{26, 27}. NHEJ proteins typically accumulate as foci of >100Kb in size around DSBs²⁸. Although, such foci are clearly visible by microscopy in single cells, the extent of damage at off-target sites appears to be too infrequent in wild-type cells to be detected in the millions of cells used by ChIP-Seq assays. Consequently, the only bona fide (>100Kb) repair focus visualized by γH2AX ChIP-Seq was at Igh²⁶. This is consistent with the fact that AID lesions at Igh are considerably more frequent than at off-target sites.

To map AID-induced DSBs across the genome the RPA-Seq technique was then developed. In this assay, AID was overexpressed from an Igκ promoter-enhancer cassette¹⁸, and the DSBs were visualized in H2AH and 53BP1 deficient B cells^{13, 14}. In the H2ax^−/− or 53bp1^−/− genetic background NHEJ is crippled, so that AID-mediated lesions occurring in G1 transit unrepaired to the S-G2M phases of the cell cycle. DSBs are then resected and exposed ssDNA recruit massive amounts of homologous recombination repair factors, such as RPA and RAD51, which are then visualized by ChIP-Seq²⁶. Under these more sensitive conditions, 235 high-confidence AID targets were identified¹⁷.

In summary, the development of TC-Seq, HTGTS, SHM-Seq, and RPA-Seq were instrumental at defining the full spectrum of AID off-targets in the mouse and human B cell genomes (see BOX 2). As discussed in detail below, recent studies from the International Cancer Genome Consortium unexpectedly uncovered a new activity for AID outside the antibody gene loci. Notably, essentially all of the AID targets identified by the new studies overlap with those catalogued by the aforementioned deep-sequencing methods.

Box 2: Deep-sequencing techniques.

To characterize AID activity across the genome, a variety of techniques based on deep-sequencing have been implemented. MutPE-Seq (mutational analysis by pair-end sequencing) measures SHM at specific genomic sites¹⁶. The domain to be interrogated is amplified by PCR and sequenced from both ends with a MiSeq 600 machine. Mutations found in both pair-end reads are considered bona fide nucleotide changes. SHM-Seq¹⁷ reveals mutations on promoter and enhancer DNA pulled down by immunoprecipitation of H3K4me3+ chromatin (ChIP-Seq). This global approach is most sensitive when repair pathways that counteract AID activity (BER and MMR) are deleted, so that AID-induced mutations accumulate over cell division. Translocations have been measured by two very similar approaches, translocation-capture sequencing (TC-Seq²⁴) and high-throughput genomic translocation sequencing (HTGTS²⁵). Both techniques rely on the constitutive induction of DNA breaks by the endonuclease I-SceI in cells carrying I-SceI restriction sites. AID-mediated or random DNA breaks that translocate to the I-SceI-mediated lesions are captured by PCR and deep-sequencing. The DNA breaks intermediate to translocations are detected by RPA-Seq. In this approach NHEJ proteins are deleted so that DSBs in G1 transit unrepaired to S and G2M stages of the cell cycle where they are resected by HR repair enzymes. ssDNA recruits massive amounts of HR factors RPA and Rad51, which are readily detected by ChIP-Seq. Finally, the potential function of AID in DNA methylation has been assessed by mapping the methylation status of AID-targeted genes in resting and activated B cells (R.C., unpublished observations). The technique used was Bi-Seq⁴³, an approach that relies on bisulfite conversion of cytidines into uracils. As methylated cytidines are resistant to conversion, genome sequencing of bisulfite treated DNA can readily discriminate between methylated and unmethylated cytidines.

Kataegis: a new AID-induced lesion?

Whole genome sequencing of a wide range of human tumors uncovered a new type of off-targeting activity for AID and related deaminases. By means of rainfall plots, which display intermutation distance per chromosome, the studies revealed multiple mutation clusters of <10Kb in size in most of the tumors analyzed^{29, 30}. Several features distinguished these ‘mutation storms’ (termed kataegis) from random substitutions, which are typically scattered across tumor genomes at a distance of ~0.1-1Mb from each other. First, kataegic mutations were predominantly (>70%) C-to-T transitions. Second, mutations at kataegis were largely fixed in the same DNA strand, a sign of catalytic processivity. Third, kataegis were frequently associated with genomic rearrangements, including localized chromothripsis, which is a phenomenon first described in cancer cells where extensive DNA damage leads to the shattering and reassembly of entire chromosomes³¹.

Based on these features, kataegis was proposed to be the result of processive cytidine deamination of ssDNA exposed by the resection of DSBs during repair³². As most mutated cytidines were found within TCX trinucleotides, the responsible factors were proposed to be apolipoprotein-B mRNA-editing catalytic subunit 3A (APOBEC3A) and APOBEC3B, which preferentially deaminate such triplets in vitro and in vivo^{29, 33}. However, in contrast to non-B cell tumors, mutations in human lymphomas derived from germinal center B cells occurred preferentially within the context of the WRCY hypermutation hotspots, providing evidence that AID is responsible for the presence of kataegis in these cells²⁹. A comparison of DLBCL and breast tumors revealed additional differences between B and non-B cell kataegis¹⁷. Whereas >80% of DLBCL kataegis were associated with transcription start sites (TSSs), only ~5% of them were found near promoters in breast cancer. Furthermore, kataegis in DLBCL were recurrent at immunoglobulin loci and mouse orthologue AID off-target sites. Conversely, in non-B cell tumors kataegis were randomly distributed across the genome, mostly at intergenic domains. AID was thus implicated in the etiology of targeted kataegis in human lymphomas, whereas APOBEC3A and APOBEC3B were proposed to generate kataegis at random sites in non-B cell tumors.

A key question is whether kataegis represent a mechanism that is distinct from and independent of SHM. Four lines of evidence suggest that this may very well be the case. One, with the exception of a few genes (e.g. BCL6), high levels of SHM are typically not observed outside the immunoglobulin loci¹². In contrast, AID-mediated kataegis are characterized by large numbers of clustered mutations near promoters of off-target genes¹⁷. Two, during kataegis, substitutions are introduced in a processive manner in the same DNA strand. Conversely, although AID can act processively in vitro³⁴, AID-mediated mutations accumulate in vivo in a stepwise manner, with only a few unlinked mutations being fixed per cell division². Three, about half of SHM substitutions are transition mutations, compared to >70% in kataegis. This disparity fits well with the proposal that kataegis likely occurs in the S phase of the cell cycle³², where uracils are mostly replicated over into thymidines or adenines (C>T and G>A transition mutations). Four, in yeast, deletion of ung1 impairs the formation of kataegis³⁵. At the same time, non-clustered hypermutation increases in the absence of ung1 in yeast cells³⁵, and in Ung^−/− B cells³⁶. This suggests that kataegis occurs downstream of DSBs, in contrast to SHM, where DNA breaks are not obligate intermediates².

Based on the above considerations it is plausible that kataegis in human lymphomas engages AID activity at two separate stages (Figure 1). Analogous to SHM, AID might first deaminate ssDNA at off-targets exposed by transcription in G1. The processing of deoxyuridine lesions by BER and MMR would then lead to occasional DSBs, which in S or G2M would be resected by the homologous recombination pathway. During or following this resection, AID may deaminate the exposed ssDNA in a processive manner, leading to long stretches of kataegic mutations. The model predicts that this second step would be independent of transcription, which is blocked following DNA damage^{37, 38}. Another important prediction is that, in contrast to AID, APOBEC enzymes should induce kataegis in non-B cell tumors by deaminating ssDNA predominately downstream of random lesions. A resolution of these questions awaits additional biochemical and genetic experiments.

Figure 1. Differential processing of AID lesions.

The deamination of genomic DNA by AID can be processed in at least 5 distinct modes. a) High-fidelity repair by BER and MMR factors can revert deoxyuridines to deoxycitidines. b) The BER and MMR pathways, in combination with translesion polymerases such as Rev1 and Polη, carry out conventional SHM. c) DNA replication over deoxyuridines in the S phase of the cell cycle leads to transition mutations. d) The formation of staggered DSBs are either repaired in cis by NHEJ leading to CSR, or in trans leading to chromosomal translocations. e) If DNA breaks are unrepaired as the cell moves to the S or G2M stages of the cell cycle, attempts to initiate homologous recombination promotes the resection of DNA ends. The resulting ssDNA are potential targets to AID, which by processive deamination may generate kataegis.

Does promiscuous AID activity play a physiological role?

Before the DNA demethylation function of TET proteins was firmly established, there was no clear mechanism driving DNA demethylation in vertebrates. Several possibilities were thus put forward, including the deamination of CpGs by AID and related enzymes³⁹. This was an attractive idea because deamination could in theory lead to the replacement of methylated cytidines with unmethylated nucleotides. However, although several reports have attempted to link AID to DNA demethylation (reviewed by³⁹), the theory faces a number of key challenges. First and foremost, AID-deficient mice do not show any obvious developmental defects, as would be predicted for a factor involved in such a critical function. Second, AID expression is confined to the B cell compartment and is not expressed by germline cells at physiological levels. Third, it is unclear how AID would access methylated CpGs in the first place, since they are by and large transcriptionally silent. Fourth, no statistically significant differences in gene expression have been reported between Aicda^−/− and wild-type B cells^40–42. Fifth, of the 235 AID targets identified in mouse B cells, only 32 (13%) are significantly demethylated during B cell activation, as measured by high-coverage Bi-Seq⁴³ (R.C., unpublished observations). Thus, a role for AID in DNA demethylation seems unlikely.

It is also important to point out that if AID promiscuous activity was in any way purposeful or had a useful role, one would expect AID targets to be evolutionarily conserved between different species. Yet, the overlap between AID off-targets in mouse and human B cells is less than 50%¹⁷. Interestingly, a recent report showed that AID is also expressed in self-reactive bone marrow B cells, and proposed that, when combined with RAGs, AID genotoxic activity might help remove autoreactive clones from the B cell compartment⁴⁴. The idea is supported on the observation that B cells treated with shRNAs against AID fail to undergo central tolerance when they recombine self-specificities in humanized mice⁴⁴. We note that under this scenario the precise genes where off-targeting damage occurs become irrelevant, so long as the damage induces apoptosis.

Thus, the available evidence so far argues against the idea that oncogenes may somehow benefit from AID off-targeting activity, but raises the intriguing possibility that the organism perhaps does employ AID promiscuity to delete autoreactive lymphocytes from the B cell repertoire.

What recruits AID activity to off-target sites?

Super-enhancers

Genomic studies that simultaneously measured DBSs, translocations, and nuclear interactions in stimulated B cells showed that interchromosomal contacts cannot explain the extent or the location of AID-induced translocations⁴⁵. Instead, oncogenes are rearranged in a manner directly proportional to the frequency of AID-mediated damage. Such findings challenged long-held ideas that Igh preferentially interacts with its translocating partners in the activated B cell nucleus⁴⁶. Thus, other properties intrinsic to off-target genes might enable the recruitment of AID activity. Several such properties were recently uncovered by analyzing AID off-targets within the context of transcription, chromatin and nuclear architecture. The key finding was that AID activity is mostly confined to super-enhancer domains^{17, 47, 48}. As discussed below, the topological and functional properties of these domains fit well with the characteristics that have long been associated with AID activity.

Super-enhancers are composed of large arrays of interconnected promoters and enhancers that display unusually high levels of transcription and epigenetic accessibility^{43, 49}. In particular, the interconnectivity between regulatory elements in super-enhancers mediates transcriptional synergy⁴⁹, consistent with the observation that promoter potency increases proportionally to the number of associated enhancers⁴³.

The interconnectivity and transcriptional synergy at super-enhancers help explain the unusual distribution of AID targets. Rather than being scattered randomly across the genome, AID-induced breaks are often found clustered into groups of 2 or 3 targets linked by long-range chromatin interactions¹⁷. Both promoter–promoter and promoter–enhancer clusters are observed, consistent with the finding that AID-mediated damage extends to transcriptionally active enhancers tethered to a targeted promoter¹⁷. The implication is that once recruited to a super-enhancer, AID deaminates topologically linked elements undergoing high levels of RNA synthesis.

Super-enhancers predominantly control transcription of genes that regulate the cell cycle and apoptosis, as well as genes that feature prominently in cell identity. These characteristics accurately describe the kinds of genes recurrently translocated in human lymphomas and mouse B cell tumors. By the same token, the transcriptional, regulatory, and architectural features of super-enhancers are shared between AID off-targets and Igh, Igκ, or Igλ loci. The key implication is that rather than mutating a specific set of genes, AID is summoned by a well-defined nuclear microenvironment, the immunoglobulin loci being its prototype (Figure 2).

Figure 2. Super-enhancers are preferred targets of AID activity.

Topological, transcriptional, and epigenetic features that render super-enhancer domains ideal targets for AID activity, both at immunoglobulin loci and at selected oncogenes in B cells, as well as in somatic cells when AID is ectopically expressed.

In support of this model, ectopic expression of AID in mouse fibroblasts leads to chromosomal translocations predominantly between genomic sites embedded within super-enhancers^{17, 48}. More importantly, the set of genes translocated in mouse fibroblasts and B cells is vastly different, owing to the fact that genes that define cellular identity (and are thus regulated by super-enhancers) differ in the two cell types. As in B cells, AID translocations in fibroblasts are associated with highly transcribed genes that display polymerase stalling at promoter areas, a key requirement for AID targeting of immunoglobulin genes⁵⁰. In addition, engagement by AID occurs at sites that share a set of epigenetic marks, including H3K27Ac and H3K36me3⁴⁸. These chromatin modifications provide nucleosome accessibility and might thus contribute to the overall targeting of AID.

An analysis that includes super-enhancers, high levels of transcription, epigenetic accessibility and interconnectivity can predict AID off-targets in a given cell type with ~90% accuracy⁴³. However, the 10% false discovery rate indicates that additional contributing factors are likely at play. Potential candidates include transcription factors, such as E2A, NF-κB and PAX5, which have long been implicated in the targeting of SHM in B cells⁵¹. However, the idea that these factors are essential for AID recruitment is challenged by at least two observations. First, as aforementioned, ectopically expressed AID targets super-enhancers in fibroblasts^{17, 48}, where B cell factors are either not expressed or are likely functionally irrelevant. Second, AID is active and appears to contribute to cell transformation when aberrantly expressed in non-hematopoietic cells⁵². Most likely then, although hematopoietic transcription factors create an AID permissive microenvironment in B cells, analogous factors must do the same in other cell types. Cell-identity transcription factors might thus recruit AID by their capacity to assemble super-enhancers.

Other factors that associate with AID and might facilitate its recruitment to super-enhancers are the RNA Pol II complex⁵³, the Pol II-associated factors SPT5⁵⁰ and SPT6⁵⁴, the Pol II elongation PAF complex⁵⁵, and chromatin modifiers, such as KAP1⁵⁶. As discussed in detail in the next two subsections, the RNA exosome complex⁵⁷ and the splicing machinery might also play a role^{58, 59}.

The RNA exosome

CSR and translocations require nicks on both strands of DNA to produce DSBs¹¹. Notably, in in vitro deamination assays, or when ectopically expressed in bacteria, AID predominantly targets the non-template strand^{60, 61}. Conversely, AID can mutate both DNA strands at sites of stalled polymerases in yeast and mammalian B cells^{59, 62}. The obvious inference is that an unknown mechanism renders the template strand accessible to AID in eukaryotes. A biochemical screen was developed to identify such a mechanism⁶². It revealed multiple subunits of the RNA exosome complex that associate with AID and facilitate SHM on both DNA strands (Figure 3A).

Figure 3. Potential mechanisms whereby the RNA exosome facilitates AID targeting.

(A) Transcription of S regions creates R-loops that cause stalling of RNA pol II, early transcription termination and RNA exosome recruitment. (B) Divergent transcription at enhancers and promoters create nucleosome-free DNA and ssDNA structures where cognate RNAs may associate and become substrates for exosome degradation. These activities are proposed to increase accessibility to AID. (C) Convergent transcription mediated by pol II may lead to the formation of RNA exosome substrates by the buildup of positive DNA supercoiling. SPT5, transcription elongation factor SPT5; TSS, transcription start site.

The exosome is a 3’-5’ RNA exonuclease complex that degrades non-coding RNAs. In the cytoplasm, it targets mRNAs that are not translated and in the nucleus it degrades prematurely terminated transcripts near promoter and enhancer transcriptional start sites (xTSS-RNAs^63–65). Compared to silent loci, sites associated with xTSS-RNAs are in general more fragile in nature, due to the frequent presence of secondary DNA structures (e.g. R-loops or G-quartets), dynamic binding of regulatory factors, extensive chromatin remodeling, and topoisomerase activity^{24, 25}. The exosome can thus be thought of as a caretaker of RNA-mediated genomic instability, by facilitating DNA repair at TSS-proximal sequences.

xTSS-RNAs are largely transcribed in a divergent orientation by polymerases moving in opposite directions^{63, 64} (Figure 3B). At enhancers, both sense and antisense transcripts are degraded by the exosome with equal efficiency, whereas antisense transcripts are preferentially targeted at promoters⁶⁶. This is due to the fact that most sense transcripts enter productive elongation and undergo early transcription termination less frequently. In addition, the exosome degrades antisense RNAs synthesized from internal TSSs at genes such as Myc and at the S domains of Igh⁶³. Since these genes also produce sense transcripts, the sites display convergent transcription originating from head-to-head TSSs⁴⁷ (Figure 3C).

The in vivo role of the exosome in relation to AID was assessed by conditional deletion of EXOSC3, one of its 11 core-subunits. Exosome-deficient B cells displayed a reduction in CSR and SHM^{63, 64}, implying that exosome-mediated degradation of Igh RNAs facilitates AID activity. The current model⁵⁷ posits that convergent and divergent transcription elicit supercoilicity on TSS-proximal DNA⁶⁷, leading to nucleosome eviction. Either the xTSS-RNAs or the exosome may then help stabilize the exposed ssDNA leading to AID attack (Figure 3). In this model it is important not to confuse, as is sometimes done, the ssDNA targeted by AID with the 8–22bp transcription bubble that is formed within the polymerase complex⁶⁸. The latter can be accessed by nucleotides during the transcription reaction but not by proteins.

In the context of CSR, the creation of ssDNA targets might be enhanced by the highly repetitive nature of S regions, which upon transcription form long stretches of ssDNA structures⁶⁹, extensive Pol II stalling^{70, 71}, and xTSS-RNA synthesis⁶³. It is important to point out however that this model does not account for AID targeting of immunoglobulin V sequences. These genes are not repetitive in nature, lack the polymerase-stalling capacity of S sequences, and do not seem to engage convergent or divergent transcription. How the exosome might then facilitate V gene hypermutation is thus unclear. Intriguingly, recent studies in ex-vivo cultures and in germinal center B cells clearly showed that V and S domains are simultaneously targeted by AID and, most importantly, at comparable frequencies⁷². Thus, the long-held notion that SHM and CSR are mechanistically or spatiotemporally uncoupled might not be correct.

Outside the immunoglobulin loci, exosome activity partially overlaps with AID off-targeting. A comparative analysis between the transcriptomes of wild-type and EXOSC3-deficient B cells reveal a genome-wide stabilization of xTSS-RNAs at a fraction of promoters and enhancers, some of which are AID off-targets^{47, 63}. Furthermore, xTSS-RNAs-associated genes are highly transcribed and often associated with super-enhancers⁶⁴. The key question however is whether these are correlative features or whether the exosome directly facilitates AID mistargeting. In support of the latter, deletion of sequences at the 5’-end of Pim1 and Cd83 TSSs reduces mutation of these genes in CH12 cells overexpressing AID⁶³.

In conclusion, the data indicate the exosome facilitates targeting of AID to both DNA strands at S domains during CSR. It remains to be determined to what extent SHM of V genes and of-target sites rely on the exosome.

Targeting AID by mRNA splicing

In this section we summarize the potential role of splicing in targeting AID to the Igh locus and potentially to off-target sites. Unlike Igκ or Igλ, Igh constant domains are independent transcriptional units, with isotype-specific promoters that drive sterile transcription of intervening (I) exons, intronic S regions, and C_H exons. S regions span 1–12kb of DNA and are composed of tandem repeats of AID-hotspot RGYW motifs. Upon transcription, S repeats form R-loops (e.g. G-quadruplexes), which enables the formation of RNA-DNA hybrids between the G-rich transcripts and the template strand^{69, 73} (Figure 4). These conformations are believed to expose or stabilize ssDNA for AID attack^{74, 75}.

Figure 4. RNA-mediated targeting of AID to switch-region DNA.

When B cells are stimulated to undergo CSR, transcription occurs at each of the recombining switch (S) regions to produce primary switch transcripts. Primary transcripts are spliced to generate a mature germline transcripts and intronic switch region transcripts (lariat intermediate). Debranching enzyme 1 (DBR1) catalyzes the release of the lariat from the spliceosome and debranches the switch transcript into its linear form. The linear switch transcript, free of exonic sequences, can function as a guide RNA by forming a G-quadruplex structure, which allows its association with AID. AID, bound to the guide RNA, is targeted specifically by sequence information provided by the guide RNAs to the complementary S region DNA.

During splicing, sterile mRNAs are partitioned into mature transcripts and lariat intronic S region transcripts (Figure 4). Early studies suggested a critical role for this splicing event in CSR, as deletion of the Iγ1 splice donor markedly reduced recombination^{76, 77}. This finding implied that either the spliced transcripts or the splicing machinery were crucial for efficient CSR. A functional link between AID and the splicing machinery has been further suggested since by the frequent coimmunoprecipitation of AID with splicing factors, including CTNNBL1, heterogeneous nuclear ribonucleoproteins (hnRNPs), nucleolin, and PTBP2^78–80. More recently, AID was also found to associate with sense S region RNA⁵⁸. Biophysical analysis showed that S region RNA folds into G-quadruplexes, and that AID associates with these structures in a glycine 133-dependent manner⁵⁸. Without the ability to bind S region RNA, AID G133V mutants fail to localize to S regions and support CSR⁵⁸. The post-transcriptional process of lariat debranching, which catalyzes the formation of linear RNA from branched spliced introns, is an additional requirement for RNA-guided AID targeting. Knockdown of the debranching enzyme DBR1 decreased the localization of AID to S regions, and mice deficient in DBR1 displayed reduced CSR. Critically, in DBR1-depleted B cells, AID localization to S regions and CSR could be rescued by the ectopic expression of S transcripts, but only in the sense orientation⁵⁸.

These findings support a model in which spliced, debranched sense S RNA forms G-quadruplexes, recruit AID, and target AID–RNA complexes back to S regions to enable CSR (Figure 4). Conversely, anti-sense S transcripts do not form G-quadruplexes and are largely dispensable for CSR⁸¹. An important implication is that unspliced transcripts either cannot form G-quadruplex structures due to steric constraints or are unable to efficiently act as an AID guide. Furthermore, the observation that deletion of the core Sµ region still allows substantial CSR⁸² suggests that the G-quadruplex RNA generated from the residual Sµ region can efficiently target AID to DNA.

A key challenge is to identify the precise biochemical mechanism that targets AID-RNA complexes to S regions. One possibility, analogous to CRISPR-Cas9 targeting, is that sequence complementarity between S region RNA and S region DNA mediates the interaction. This scenario would require either the displacement or the collapse of R-loops, because S region RNA-DNA hybrids are exceptionally stable and unlikely to be outcompeted for access to S region DNA⁷³.

IgV regions and most gene transcripts do not form G-quadruplexes or R-loops. Hence, a link between splicing and SHM or off-targeting is unclear. It is also noteworthy that A:T-rich S regions from Xenopus laevis can serve as CSR substrates when inserted into mouse B cells, suggesting that additional mechanisms can target AID to these regions⁸³. Nevertheless, mRNA splicing might play a role in AID mistargeting of at least a subset of genes. For instance, some reports have correlated AID activity at non-immunoglobulin loci to G-richness⁸⁴. Furthermore, transcription of the immunoglobulin-translocation partners MYC, BCL6 and RHOH in lymphoma cells appears to induce the formation of G-loops⁸⁵. This raises the possibility that G-quadruplexes, whether at the DNA or RNA level, could enhance AID accessibility to oncogenes and other loci. The G133V AID mutant, which cannot localize to S regions, provides an opportunity to explore such ideas.

In summary, the recent findings cement a role for mRNA splicing in AID targeting during CSR. It now remains to be determined whether this mechanism also facilitates SHM of V genes or off-targeted oncogenes.

Keeping AID off-targeting activity at bay

B cells have developed a plethora of mechanisms that tightly control AID mRNA and protein abundance, nuclear access, preferential targeting to immunoglobulin loci, and catalytic activity. In addition, DNA repair pathways eliminate DSBs and off-targeting SHM and actively remove cells bearing AID-induced translocations. The picture emerging from the available data suggests that these mechanisms ensure an optimal equilibrium between the ability to mount efficient antibody responses and the risk of oncogenic transformation, but do not play a major role in defining the specificity of AID targeting.

Transcriptional and post-transcriptional regulation

In mammals, Aicda transcription is maximal in activated or germinal center B cells, where T cell-derived cytokines (e.g. IL-4 and pro-inflammatory cytokines) promote transcription factor binding to upstream regulatory elements within the Aicda locus^{4, 43, 86–90}. Aicda is also induced by T cell-independent factors, including BAFF, APRIL, Toll-like receptor (TLR) signaling, and female sex hormones^{87, 91–93}.

Both general and B cell-specific transcription factors activate Aicda, including NF-κB, PAX5, STAT6, IRF4, C/EBP, E-proteins and FOXO1^{89, 90, 94–97}. Conversely, ID2 and BLIMP1 repress Aicda and promote B cell differentiation to the plasma cell stage^{87, 96, 97}. Inflammatory cues also promote ectopic Aicda expression in non-germinal center B cells upon viral infection^{87, 98}.

Post-transcriptionally, Aicda expression is regulated by the microRNAs miR-155 and miR-181b, which downmodulate AID protein levels and activity^99–101. Conversely, in AID⁺ chronic myeloid leukaemia and acute B lymphoblastic leukaemia cells, miR-155 is not expressed, a feature that may exacerbate the pathological role of AID in those malignancies¹⁰².

The extent of antibody diversification and chromosomal translocations are directly proportional to Aicda expression levels. In Aicda^+/− mice, SHM, CSR and chromosomal translocations are reduced approximately by half^{103, 104}. Correspondingly, B cells lacking the miRNAs that regulate AID display increased CSR but also tumor-inducing translocations^{99–101, 105}. Interestingly, although enforced overexpression of Aicda can be oncogenic¹⁸, AICDA levels in cancer cells are usually similar to or lower than those in normal B cells (reviewed in REF. ¹⁰⁶). This suggests that AID off-targeting does not require overexpression. Indeed, low Aicda expression in pre-B cells or normal AID levels in germinal center B cells are sufficient to generate oncogenic lesions^{107, 108}.

Protein compartmentalization and stability

AID is a nuclear–cytoplasmic shuttling protein^109–111. Under steady state conditions, ~90% of AID is cytoplasmic, where its half-life is substantially longer than in the nucleus¹¹². The mechanisms that regulate AID subcellular localization and protein stability establish a dynamic equilibrium that might help reduce AID pathological activity while allowing efficient antibody diversification.

In the cytoplasm, AID is stabilized by the HSP90 chaperone pathway¹¹³, which includes HSP90 itself and the farnesylated co-chaperone DNAJA1, one of 42 DnaJ proteins in vertebrates¹¹⁴. The HSP90–DNAJA1 complex regulates AID protein levels for optimal antibody diversification. Indeed, activated B cells from DnaJa1^−/− mice exhibit a 50% reduction in AID protein levels and a proportional decrease in SHM and CSR¹¹⁴. Similarly, pharmacological inhibition of HSP90 results in a dose-dependent decrease in AID protein and activity^{113, 115}. As such, these inhibitors could be valuable to reduce AID-dependent clonal evolution in B cell malignancies ¹¹⁵.

In addition to HSP90, cytoplasmic AID exists within a high molecular weight complex containing the translation elongation factor EEF1A¹¹⁶. The AID–HSP90 and AID–EEF1A complexes are physically and functionally distinct in that they both regulate AID half-life by different pathways¹¹⁷. HSP90 stabilizes immature AID, while EEF1A keeps AID out of the nucleus^{113, 117}, where AID is actively targeted to the proteasome via the adaptor REGγ and an unknown E3 ubiquitin ligase^{112, 118}. AID shuttling in and out of the nucleus is controlled by at least three mechanisms: one, a nuclear export signal (NES) at the C-terminal of AID is recognized by the exportin CRM1, which shuttles proteins and RNAs out of the nucleus in eukaryotes¹¹⁹. The association of AID with CRM1 is likely regulated by the RAS-related nuclear protein (RAN)–GTP/GDP differential on either side of the nuclear membrane. Irreversible inhibition of CRM1 by leptomycin B moderately increases the proportion of nuclear AID^{109–111, 117}. Two, AID’s molecular mass of 24 kDa is below the maximum size for passive nuclear diffusion through the nuclear pore. In spite of this, AID is purposely imported into the nucleus by a process that probably involves karyopherin α and karyopherin β^{79, 120}. Three, a proactive nuclear import is required for AID because it is sequestered in the cytoplasm by a multiprotein complex containing a transfer RNA-free version of EEF1A, which functions in a manner unrelated to its role in protein biosynthesis^{116, 117, 120}. AID interacts with the EEF1A complex via a domain that partially overlaps with, but is distinct from, its NES. The stoichiometry of the association of AID with EEF1A suggests that this complex is the major cytoplasmic reservoir of AID¹²¹. Molecular modeling and structure–function analysis suggest that the EEF1A–binding motif and the NES of AID reside on opposite sides of an amphipathic helix. If confirmed, this feature explains how the small C-terminal domain of AID mediates both cytoplasmic retention and nuclear import^{117, 120}.

Despite its accumulation in the cytoplasm, AID does not seem to play a role in this compartment. However, it is possible that this partitioning is a relict of an ancient AID role against viral infections, played now by the APOBEC enzymes, which evolved from AID¹²². Cytoplasmic retention might have later evolved to minimize AID off-targeting activity in the nucleus.

Post-translational modifications

AID is phosphorylated in at least five residues. PKA phosphorylates AID at Ser38 apparently only within the context of Igh chromatin^{123, 124}. This modification enhances AID enzymatic activity during CSR and SHM^123–125. An obvious implication of constraining AID Ser38 phosphorylation to the Igh is that it might help prevent the formation of chromosomal translocations, although this idea has not been directly tested. Three other phosphorylation events at Ser3, Ser27, and Thr140 have been shown to either inhibit or enhance AID enzymatic activity, but only Thr140 has a known physiological role so far, in that it potentiates SHM and CSR^126–128. Finally, Tyr184 at the C-terminal of AID is phosphorylated in vivo but no role in AID compartmentalization or activity is apparent¹²⁹.

Figure 5 integrates the various mechanisms that regulate AID expression and activity into a single model. HSP90 stabilizes metastable AID in the cytoplasm, probably until AID adopts a conformation that permits its association with the EEF1A complex^{113, 117}. Nuclear export and cytoplasmic retention act in parallel to exclude AID from the nucleus¹¹⁷. Disruption of the AID–EEF1A interaction facilitates CSR but also increases the likelihood of chromosomal translocations¹¹⁷. Inhibiting HSP90 reduces both antibody diversification and off-target effects^{113, 115}, whereas releasing AID from EEF1A increases both CSR and translocations¹¹⁷. These conflicting observations might reflect the fact that when associated with HSP90, AID has not yet acquired a stable conformation, whereas the opposite is true when AID is complexed with EEF1A¹¹⁷. Similar to transcriptional regulation, neither of these mechanisms controls AID targeting specificity but rather they determine the magnitude of its activity. Intriguingly, CRM1 inhibition does not affect CSR, despite increasing AID nuclear levels¹¹⁷. Thus, nuclear export and cytoplasmic retention may not be functionally equivalent. Once in the nucleus, AID is destabilized by REGγ and ubiquitylation^{112, 118}. REGγ-deficient B cells display increased levels of nuclear AID leading to higher CSR but, interestingly, they do not display an increase in chromosomal translocations¹¹⁸. This is in contrast to inhibition of EEF1A and may reflect the fact that REGγ regulates AID at a latter step, perhaps following targeting of the immunoglobulin gene loci, whereas when released from EEF1A AID is capable of targeting the entire genome.

Figure 5. Integrative scheme of the mechanisms that regulate AID activity.

AID transits through HSP90- and eEF1A–containing complexes in the cytoplasm that control its functional maturation before being imported into the nucleus. Whether nuclear import of AID is regulated or stochastic is unknown. In the nucleus AID has multiple possible destinations. It can be: one, exported back to the cytoplasm, two, destabilized by proteasomal degradation, three, associate with the nucleolus and four, be recruited to the chromatin of immunoglobulin variable (IgV) and/or S regions as well as to off-target genes. It remains to be determined to what extent these various destinations are sequential while AID travels to the Ig loci, or compete for AID or synergize to minimize off-targeting. Targeting of AID to the IgV or S-regions requires strong transcription by RNA polymerase II associated with Spt5 and the PAF complex in the context of a specific chromatin microenvironment that includes abundant topological associations of multiple enhancers. Transcription can lead to DNA supercoiling and expose ssDNA that is stabilized by RPA. The repetitiveness of S-regions facilitates the formation of R-loops, exposing the untranscribed DNA strand. The exosome helps recruit AID to the template strand of R loops by degrading R-loop associated transcripts. AID is also recruited by genomic regions displaying extensive convergent or divergent transcription. A number of chromatin associating factors might help tether AID to chromatin, at least at S regions. These include KAP1, 14-3-3 and PTBP2. Once recruited, AID initiates a series of events leading to the formation of DNA breaks (after excision of uracil by UNG and the DNA nicking activity of the endonuclease APE1), which activates the DNA damage response kinase ATM. ATM promotes PKA-mediated phosphorylation of AID at serine 38. This phosphorylation increases AID activity and allows AID association to RPA. In addition, AID binds to G-quartet structures formed in the debranched intron of the S-region sterile transcript, which presumable pairs back with the DNA and helps targeting AID for CSR. Black arrows indicate mechanisms thought to act in chronological order and/or that are directly linked. Dashed grey arrows indicate possible but not yet empirically demonstrated connections.

DNA repair mechanisms both curb and facilitate AID pathological activity. First, as previously discussed, BER and MMR pathways reverse off-targeting hypermutation and the extent of this activity appears to differ depending on the genomic locus¹². At the same time, it should be noted that BER and MMR generate DNA nicks and gaps, which are either filled in by DNA polymerases to allow for the full spectrum of SHM, or result in DSBs that promote CSR and chromosomal translocations^{21, 130}. Finally, multiple components of the DNA damage response, including ATM, NBS1 and p53 eliminate cells carrying AID-induced chromosomal translocations¹³⁰. This mechanism does not regulate the targeting specificity of AID but is key to prevent AID-induced B cell lymphomas ¹⁸.

In conclusion, the accumulation of AID in the B cell nucleus is controlled by the interplay between nuclear import and export, cytoplasmic retention, AID protein turnover, and possibly cell cycle regulation (Box 1; FIG. 5). DNA repair pathways provide an additional layer of control. The expectation is that the cross-talk between these regulatory pathways ultimately dictates the extent of SHM and CSR, as well as off-target deamination and damage. However, it is important to point out that these mechanisms limit the magnitude but do not control the specificity of AID off-targeting activity.

Box 1: Regulation of AID activity during the cell cycle.

AID-mediated deamination and its processing by BER or MMR is constrained to the G1 phase of the cell cycle^{131, 132}. Likewise, downstream DSBs appear to be rapidly repaired in G1^{26, 133, 134}. B cells lacking factors necessary for NHEJ or homologous recombination accumulate AID-mediated breaks in S and G2/M^{26, 135}, These lesions might represent delayed repair or in some cases active deamination at S and G2M. Some mechanisms that might restrict AID activity to G1 include stage-specific regulation of AID protein localization and/or stability^{136, 137}, phosphorylation, availability of co-factors, or the regulation of nuclear AID stability by cyclins, as has been recently proposed¹³⁸. To date, this topic remains largely unexplored.

Moving forward

Driven by advances in genomics, AID off-targeting activity has been narrowed down in the past few years to a nuclear microenvironment characterized by the presence super-enhancers, extensive interconnectivity between regulatory elements, high levels of convergent and divergent transcription, and a high level of RGYW accessibility. The remaining task is to elucidate the factors rendering these microenvironments preferred AID targets. A large number of co-immunoprecipitation and genetic studies have consistently implicated three general mechanisms in AID activity: splicing, transcription (mediated by Pol II-associated proteins and transcription factors), and RNA degradation. However, demonstrating the role of the isolated factors in AID targeting has been complicated by the fact that such activities are often essential to cell proliferation, which is in turn required for AID activity. Furthermore, knockdown experiments are by definition incomplete and do not always replicate under different culture conditions or in conditional knockout settings. Thus, new techniques that enable acute and transient depletion of proteins of interest are needed to overcome such shortcomings. Solving these challenges will be key to achieving a full understanding of the mechanisms that facilitate the high affinity of AID for antibody genes, and its lesser but pathological off-targeting of selected oncogenes.

Glossary Terms

V(D)J recombination

Somatic rearrangement of variable (V), diversity (D) and joining (J) regions of the genes that encode antibody and T cell receptor proteins. The combinatorial nature of V(D)J recombination and the distribution of recombining genes in the vertebrate genome creates repertoire diversity of T and B cell surface receptors

Somatic hypermutation (SHM)

A unique mutation mechanism that is targeted to the variable regions of rearranged immunoglobulin gene segments. Combined with selection for B cells that produce high-affinity antibody, SHM leads to affinity maturation of B cells in germinal centres

Class-switch recombination (CSR)

A recombinational process that replaces the immunoglobulin heavy chain constant region Cµ (which encodes the Fc portion of IgM) for that of a downstream isotype Cγ Cα or Cε which encode the constant region of IgG, IgA and IgE, respectively

DNA deamination

Removal of an amine group from pyrimidine or purine nucleic-acid bases. Deamination of cytosine and adenosine yields uracil and inosine, respectively

Base-excision repair (BER)

A DNA-repair pathway that removes uridine nucleotides from DNA, that arise by spontaneous of purposeful deamination of cytidines. Repair is initiated by the DNA glycosylase UNG that excises the uracil base, followed by cleavage of the abasic site by the apurinic apyrimidinic endonuclease 1 (APE1)

Mismatch repair (MMR)

A repair pathway that removes mismatched base pairs from DNA that result from errors made by replicative DNA polymerases or from deamination by AID and APOBEC deaminases. Repair involves the removal of a tract of DNA including the mismatch, and re-copying of the complementary strand. This pathway is mediated by proteins MSH2 and MSH6 among others

Chromosomal translocations

Aberrant joining of DNA breaks from heterologous chromosomes that do not normally pair during mitosis or meiosis

AID off-targeting activity

Promiscuous AID-mediated cytidine deamination of genomic sites other than immunoglobulin gene loci

Enhancer

A regulatory DNA element that recruits transcription factors and influences the rate of gene expression. Enhancers function in an orientation- and position-independent manner (that is, they can function either upstream or downstream of the associated gene, or in an intron). They believe to associate to promoters via long-range chromatin interactions

Non-homologous DNA end joining (NHEJ)

A repair pathway that joins broken DNA ends without depending on extended homology. Components of this pathway include the proteins Ku70, Ku80, ARTEMIS, X-ray repair cross-complementing protein 4 (XRCC4), DNA ligase IV, and the catalytic subunit of DNA-dependent protein kinase (DNA-PKcs)

Kataegis

Clusters of mutations (mostly transitions) in the same DNA strand introduced in tumor genomes by cytidine deaminases: APOBEC enzymes in non-B cell tumors and AID in B cell lymphomas

Chromothripsis

Clustered and massive chromosomal rearrangments in one or several chromosomes of primary or transformed cells. This process occurs as a result of a catastrophic event in the history of the cells and promotes tumor development and congenital diseases

Transition mutations

Base changes in DNA in which a cytidine (C) or thymidine (T) is replaced by a T or a C, respectively. A to G and G to A mutations are also transitions

Homologous recombination

DNA repair pathway that makes use of homologous sequences (e.g. homologous chromosomes) as template to repair a DSB. The process involves resection of DNA ends, recruitment of RPA and Rad proteins, strand invasion of the intact sequence, DNA synthesis, ligation, and resolution

TET proteins

A family of proteins that catalyze the conversion of methylated cytidines to hydroxymethylated cytidines. This step initiates a series of catalytic events that leads to DNA demethylation

Super-enhancer

A cluster of transcriptional regulatory elements (promoters and enhancers) associated by long-range chromatin loops. They tend to modulate gene expression as a unit

RNA exosome complex

Multiprotein intracellular complex that degrades short RNA molecules in the 3’-5’ orientation

xTSS-RNAs

Transcription start site (TSS)-associated antisense transcripts that can exceed 500 base pairs in length and are transcribed divergently from cognate coding genes. These RNAs are mostly degrated by the exosome complex

G-quadruplexes

Non-B DNA structures that form at G rich sequences. By means of Hoogsteen hydrogen bonding guanines create square planar structures known as tetrads. Two or three tetrads can stack on top of each other to form a quadruplex