DNA Acrobats of the Ig Class Switch ¹

Immature B lymphocytes mature into small resting B lymphocytes with an Ag receptor of the IgM class. Without prior interaction with Ag, B cells join gene segments to create the Ag-binding Ig V region. Because numerous gene segments can be joined in numerous different combinations, the naive B cells are able to create a sizable repertoire of Abs. Yet the Abs of this primary repertoire are of a lesser quality; it seems reasonable that in any large repertoire not aimed at a particular Ag, low-affinity Abs should far outnumber high-affinity Abs. However, the low affinity is in part offset by the higher avidity of the IgM Abs that are secreted by the plasma cells derived from the activated B lymphocytes. In the secreted IgM, the prototypical four-chain Ig structure is combined into pentamers or hexamers. In this case, the trade-off for higher avidity is a clumsy structure that does not penetrate the entire organism; particularly, IgM does not pass the placenta and thus cannot protect the fetus. Such protection depends on a switch from IgM to IgG during the course of an immune response. In more general terms, switching to other Ig classes serves to distribute a particular V region onto different effector functions and guide it to different sites in the organism. Some 30 years ago, it was shown that class switching occurs within a committed (i.e., surface Ag receptor-expressing) lymphocyte (1, 2, 3), which in turn required an explanation for how a given V region could be grafted onto various C regions at the genetic level. The basic principles of how this happens have been elucidated over the last three decades, and the biochemical details are now becoming clearer.

As mentioned above, the primary Ab repertoire is manifested in IgM Abs, and cooperative binding by the 10 or 12 identical Ag combining sites presumably counterbalance low affinity to some extent. However, because the other secreted Ig classes have only two (IgG, IgA, IgE) or four (IgA) combining sites, this compensation is lost when the IgM H chain switches to the other classes. Here, affinity maturation comes to rescue, and Ab quality improves, often dramatically, upon exposure to Ag. First, the clones of the primary repertoire with higher affinity to the Ag are selected and induced to proliferate. During this proliferation, mutations arise in the Ab V regions at a rate that is a million-fold higher than the normal, spontaneous mutation rate (somatic hypermutation). If the mutations result in Abs of even higher affinity, these clones are then selected, and they proliferate further. The temporal connection between class switching and hypermutation is also a mechanistic one, because both are initiated by activation-induced cytidine deaminase (AID) ³ (4, 5, 6) and mediated, in part, by uracil-DNA glycosylase (UNG) (7, 8). Although this review will focus on class switching, it will take some key observations from the hypermutation process to shed light on the mechanism of class switching.

Soon after it was recognized that the joining of V region gene segments requires DNA deletion, it was discovered that the genes encoding the H chain isotypes other than μ or δ are also generated by joining and deletion (9). This deletion occurs between so-called switch (S) regions—repetitive sequences 5′ of the constant (C) gene segments (10). These C gene segments are arranged in a linkage group downstream of the V exon. During switch recombination, the Cμ (which encodes the C region of the H chain of IgM) is replaced by the C region of another isotype. A priori, such a recombination process must be mediated by double-strand breaks that are subsequently ligated (repaired).

When the DNA is cut at two places this leaves four new ends. These ends may be rejoined in three different ways. One restores the original configuration. Another way leads to an inversion of the DNA between the cuts. The third way leads to deletion; the intervening deleted sequence ligates its ends to form a circle (switch circle), which is lost. The remaining ends are rejoined, grafting a different C region with the extant V region. If the ends were free and not in proximity to each other when the cuts are rejoined, then circular religation (a unimolecular reaction) and deletion should be highly favored over inversion (requiring bimolecular reaction). But because inversions are found almost as frequently as switch deletions (11), this implicates more advanced DNA acrobatics. We and others have predicted the existence of switch loops (11, 12, 13, 14) where the DNA ends are in proximity before cutting and rejoining (Fig. 1). Looping-out could occur after or before the first cut, but needs to precede the second cut in order for the inversion scenario to occur. Because a switch recombination event requires two breaks separated by long distances (up to 175 kb), a looping mechanism likely would make the rejoining of the chromosomal ends efficient.

Although the looping of the DNA between S regions could occur by the spontaneous, random motion of the DNA, we and others have sought more active switching components. Here the formation of the switch loop could be facilitated by components of a switch recombinase system. Such a system would be able to recognize and bring together the S region of μ (Sμ) and the S region of the future isotype. Although the repetitive sequences of the S regions share some homology, the presence of inversions indicates that switch recombination is not mediated by homologous recombination. The fact that inversions between the S regions can occur without homology at the joining sites (11, 15) further implicates nonhomologous recombination.

Thus, switch recombination uses components of the nonhomologous recombination pathway for general DNA repair. Clearly involved in the switch process is the histone H2AX (16). In its phosphorylated form, it helps prevent aberrant repair of both programmed and general DNA breakage by recruiting the Mre11-Rad50-Nbs1 complex to the repair foci. Mice that lack either one or both H2AX alleles in the concomitant absence of p53 develop B lineage lymphomas with chromosome 12 (IgH)/15 (c-myc) translocations, with hallmarks of either aberrant V(D)J or class switch recombination (17) (however, H2AX is not needed for hypermutation (16)). But, although B cells lacking H2AX show impaired authentic class switch recombination, the frequency of intraswitch region deletions is normal (16). This indicates that lesions and ligation are normal, but long-distance joining is not. Thus, histone H2AX may facilitate synapsis by chromatin remodeling.

Another candidate for attachment to the S regions and loop formation are the Ku70 and Ku80 proteins of the nonhomologous end joining repair pathway, because both are required for switch recombination (18, 19). Working together as a heterodimer (20), Ku70/Ku80 binds to ss- and dsDNA ends. It has been suggested that Ku70/Ku80 can bind to some DNA sequences without open ends (21). Perhaps the (predicted only) proclivity to form hairpin-like structures avails the S regions to Ku70/Ku80 or other factors binding (21) before DNA cuts occur. These secondary structures would occur within a particular S region; it is different from the looping out of dsDNA, which is deleted as a result of a completed recombination process. So we can imagine that one or more proteins, perhaps Ku70/Ku80, bind to Sμ and to another S region and bring them into close proximity. This might occur with (22) or without (23, 24) the help of DNA-PKcs. In conventional (nonhomologous) double-strand repair, DNA-PKcs is a DNA-dependent phosphokinase that is recruited by Ku70/80. For class switching, however, there is no absolute requirement for DNA-PKcs (see Ligating DNA ends).

The intrastrand secondary structure by itself may not be sufficient for the S region to be a target of the recombinase. As with hypermutation, switching requires transcription (Fig. 2) (Refs.25, 26, 27 ; reviewed in Ref.28). This is in line with the activity of AID, which apparently requires ssDNA exposed by transcription or by specific binding proteins. This requirement was also demonstrated by the transcription-dependent recombination of an exogenous substrate in AID-producing fibroblasts (29). In particular, switching requires synthesis of a germline transcript that is composed of the I exon, the intron containing the S region, and the C region gene segments (Fig. 2,A). When Sα is transcribed in the physiological direction in vitro, stable RNA-DNA structures could form, and the newly synthesized RNA molecule would remain associated with the S region DNA template strand (30, 31). In vivo, the so-called R-loop (Fig. 2,B) can exceed 1 kb in length (32). This would expose the nontemplate DNA strand for putative switch recombinase factors. However, for class switching to occur, not only must there be transcription from the I exon promoter, but also the transcript must be spliced (33, 34). Perhaps splicing factors need to be recruited to the S regions (35). Alternatively, the spliced transcript could hybridize to the S region DNA, allowing it to form a ssDNA switch bubble (Fig. 2 C). It is also possible that the spliced transcript interacts directly with AID. It would be somewhat surprising if a minimal R-loop model (i.e., a transcription bubble) would explain the great efficiency of switching after B cell activation. If only a transcription bubble would be needed, perhaps many other transcribed genes might recombine.

The switch recombination breakpoints in frogs (36) and mice (37) are usually at positions (microsites) where a ssDNA folding program predicts the transition from a stem to a loop structure within the S region. It seems clear that AID binds to the chromatin at the S region (38) and, in conjunction with UNG (7, 8, 39) and yet-to-be-identified apyrimidinic endonuclease(s), introduces nicks into DNA (Fig. 3). This model suggests how AID cuts DNA—many nicks in close proximity would lead to the eventual double-stranded break (40, 41), which are staggered (42). Indeed, AID is required for DNA cleavage in switch recombination (43), although it has also been concluded that AID functions in a postcleavage step, at least in the somatic hypermutation process (44). Specifically, AID may (4) or may not (45) deaminate free cytidine, but it converts also deoxycytidine into deoxyuridine in DNA (41, 46, 47, 48, 49). The noncanonical DNA base deoxyuridine, in turn, may be removed by UNG (7, 8), thereby creating an abasic site on one DNA strand. This site then may be attacked by an apurinic apyrimidinic endonuclease, which creates the ssDNA breaks.

In the hypermutation process, when one analyzes DNA sequences, it appears as if AID preferentially attacks a G on the reading strand, embedded in an RGYW motif (with R representing a purine; G, deoxyguanidine; Y, a purine; and W, either A or T) (50, 51). In fact, we know that AID attacks a C in ssDNA (41, 46, 47, 48, 49); therefore, the G in the short motif marks the site of attack of the C on the other DNA strand. Because S regions have many repeats of AGCT—a palindromic RGYW motif—there will be many mutations (16), and, as a consequence, many nicks introduced into the S region DNA. Moreover, the G·U mismatch may be dealt with by mismatch repair, which removes longer stretches of ssDNA around the mismatch by an exonuclease (52) (Fig. 3); if this stretch happens to have a single nick on the other strand, a double-strand break would also result. In fact, components of the mismatch repair system are involved in the switch process (53, 54, 55, 56, 57). Without them, there is less switching; and the recombination breakpoints are more likely to occur in consensus motifs.

The sequence of events described above do not quite fit all of the characteristics ascribed to AID. Some experiments have shown that AID introduces C-to-U mutations at the nontemplate strand only (47). This means that nicks would be introduced in the upper (reading) strand only, and no double-strand breaks could be formed without spontaneous nicking on the template strand. Also, exonuclease activity around the nick would not help under those circumstances—the other strand would simply not be nicked. However, it is possible that repair is subverted in that it works on the nonmutated rather than on the mutated strand, which would introduce the lesions on the other strand. A similar role of mismatch repair has been suggested to contribute to hypermutation (58). At any rate, the preferred AID target of a G on the reading (upper, nontemplate) strand embedded in the RGYW in hypermutation is not consistent with C being targeted in the nontemplate strand. Nor is the fact that RGYW apparently can be targeted on both strands in cultured cells (59, 60, 61) and in the whole animal (62).

It is also known that the requirements for AID structure on class switching are greater than on hypermutation. If mutated, the C-terminal end of AID still can mediate hypermutation, but not class switching (63, 64). Apparently, the C-terminal region needs to bind another molecule, which may be another protein, but also DNA, or RNA. There are also cell lines that express endogenous AID, but do not switch extrachromosomal switch substrates (65); and there is another one that does not switch or hypermutate the endogenous μ gene. ⁴ Finally, mitogen-stimulated B cells switch but do not hypermutate their Ig genes (66).

In the work of Ito et al. (67), it has been proposed that AID, in addition to working on DNA, also edits a specific RNA. At the moment, however, there is no experimental evidence that AID can in fact act on RNA, let alone any evidence that it does act on RNA physiologically. But it is quite interesting that AID is a nucleocytoplasmic shuttling protein with a bipartite nuclear localization signal and a nuclear export signal (67), although it is predominantly found in the cytoplasm when tagged by green fluorescent protein (68). It has been suggested that AID may have mechanisms similar to those of apolipoprotein B mRNA-editing catalytic polypeptide 1 (APOBEC1) (67). According to this hypothesis, edited mRNA (encoding an unknown polypeptide) in the AID-cofactor complex is transported to the cytoplasm for translation. The amino acid sequence of AID is somewhat similar to APOBEC1, and when AID was initially discovered, it was suggested that AID would also be an mRNA-editing enzyme. If so, it could prove wrong most of the DNA cutting pathway proposed earlier in this review. With the RNA-editing mechanism of APOBEC1, this cytidine deaminase introduces a premature stop codon into apolipoprotein B100 mRNA. The majority of these edited mRNA molecules is exported to the cytoplasm as a complex associated with APOBEC1 and a cofactor, APOBEC1 complementation factor, to avoid nonsense-mediated decay. Nonsense-mediated decay affects mRNAs with premature stop codons, presumably to avoid translation of short polypeptides that would act as dominant-negative factors. The similarity between AID and APOBEC1 activity might hinge on the inhibition of nonsense-mediated decay, which works quite efficiently in B lymphocytes (69). The I exon of the germline transcripts contains stop codons in all three frames. Without protection from nonsense-mediated decay, there would not be a sufficient quantity of transcripts to form the switch bubble. Perhaps AID and the putative cofactor need to bind to the spliced germline transcript. Interestingly, when synthesized in insect cells, AID is bound to RNA, which needs to be removed by RNase for AID to exhibit its enzymatic activity (48). Presumably, nonspecific insect mRNA blocks binding of specific nucleic acid. However, the fact that spliced germline transcripts cannot be supplied in trans (33, 34) would speak against the course of events outlined above.

After the appropriate cuts are made, the open DNA ends need to be ligated, i.e., the lesions need to be repaired. Indicative of an involvement of the Mre11-Rad50-Nbs1 complex, B cells from patients with Mre11 deficiency (ataxia telangiectasia (AT)-like disorder) switch less efficiently, and the switch recombination junctions are aberrant (70). In the main pathway for nonhomologous end joining, the phosphokinase DNA-PKcs, after it has been recruited by Ku70/80, phosphorylates itself and Ku70/80, and colocalizes at DNA damage foci with other repair proteins such as H2AX and 53BP1 (71). The Ku/DNA-PKcs complex also recruits and phosphorylates XRCC4 and ligase IV (reviewed in Ref.20). The requirement for DNA-PKcs in switch recombination is not absolutely clear. On the one hand, it seems natural that, with Ku70/80 being strictly needed, DNA-PKcs would follow. In cultured cells, such a requirement was reported (22). Yet, in the mouse, this requirement is not strict (23), particularly not for the IgG1 isotype (24). Perhaps ATM (reviewed in Ref.72), the protein mutated in AT, can replace DNA-PKcs (73). ATM is another important protein kinase in the double-strand repair pathway; and AT patients suffer from immunodeficiency. Whatever the exact requirements for kinases, in the end, ligase(s) need to join the open ends to generate the complete switched gene, in which the V exon is now in close proximity to the C region gene segment of the new isotype. To date, the identity of the ligases required are not known, although there are indications that ligase IV is among them (74).

At the cellular level, B cells need input from Th cells in order for them to switch their Ig isotype (reviewed in Ref.75). There is a large body of evidence of how different lymphokines instruct the switch recombinase system to a particular isotype (reviewed in Ref.28). The lymphokines are synthesized by the Th cells, which in turn have been directed to differentiate into Th1 or Th2 cells by lymphokines synthesized by professional APCs. Different types of pathogen are handled by different APCs, which in turn synthesize different lymphokines. This is why in the end, the pathogen instructs the isotype of the Ab. However, the question remains as to how the lymphokines accomplish this. Obviously, as they bind to the appropriate receptor on the surface of the cell that is to switch Ig isotype, a signaling cascade is initiated that ultimately leads to transcription off the I region promoter (reviewed in Ref.28). But is this and AID all that is needed to set off the switch process? In other words, are there no S region-specific factors needed (76)? S regions do differ among each other (77), and they differ more from one S region to another than between the same S region of different species (78, 79). This may indicate other requirements (65, 80), in addition to transcription and subsequent splicing. Various factors that bind to S regions have been reported; however, none of these factors are expressed specifically in switching B lymphocytes. Some of these factors are well characterized in processes other than switch recombination. They include Pax-5/B cell-specific activating protein (reviewed in Ref.81), LR1 (82), NF-κB/p50 (83, 84), switch nuclear A-site protein (includes E47) (85, 86), and SμBP-2 (87). Among them, NF-B/p50, Pax-5, and E47 have also been shown to affect levels of Ig switching, with the E proteins directly regulating expression of AID (88). Recently, a ubiquitously expressed, DNA-binding, late SV40 factor has been reported to bind S regions and to repress class switching to IgA (89). However, in general, the direct role of these factors in the switch recombination process has yet to be integrated. But, clearly, there may be more factors needed than the well-studied ones. Indeed, there are hyper-IgM syndrome patients with a selective defect that has not been attributed to any of the known genetic defects (90). In some of these patients, the defect might lie in survival signals delivered to switched B cells, but in others, switching might be impaired.

The S regions seem to be particularly good targets for recombination because of their highly repeated WRCY (RGYW on the complementary strand) sequence motif, where an embedded C can be a target of AID. Apart from transcription, subsequent splicing of the germline transcript is necessary in order for the switch recombination to proceed. No such specific transcript has been identified for hypermutation of the V region—a process that shares with class switching the requirement for AID, UNG, and mismatch repair proteins. It almost seems as if hypermutation at the V region is a byproduct of the class switching; with affinity maturation as a welcome afterthought when the switch process leads to the loss of avidity. Although hypermutation results in a much higher frequency of mutations, the intrinsic rate, with 10⁻³ to 10⁻⁴ per base pair per cell generation (91, 92, 93), pales in comparison to another mutation process mediated by a relative of AID-APOBEC3G, which inactivates retroviruses. APOBEC3G incapacitates spreading of retroviral infection by mutating the minus cDNA strand (94, 95), which is synthesized with genomic viral RNA as a template. In one round of replication, up to 13% of the G bases can be mutated (95). The lower activity of AID seems to be compensated by the abundance of RGYW target motifs in the S regions.

The interest in the basic mechanism of Ig class switching is also fed by its immediate clinical impact. IgA deficiencies are relatively common. Even more important is IgE as the primary culprit in immediate type of hypersensitivity. IgE-secreting plasma cells have also been implicated in human asthma. Finding the hypothetical factor that guides the switch recombinase to generate the gene encoding the ε-chain of IgE may provide the basis for a simple screening test, in which chemicals are assayed for their potential to bind to the factor and, thereby, to inhibit its function. Chemicals that inhibit the factor will inhibit the generation of IgE, and thus have the potential to prevent an allergic reaction to any substance.