APE2 Promotes AID-Dependent Somatic Hypermutation in Primary B Cell Cultures That Is Suppressed by APE1

Somatic hypermutation (SHM) and Ab class-switch recombination (CSR) are both initiated by activation-induced cytidine deaminase (AID) and involve error-prone DNA repair. SHM occurs in germinal center (GC) B cells, and whereas CSR can occur in activated, cultured B cells and a few B cell lines, SHM does not. We previously reported differential expression of the apurinic/apyrimidinic (AP) endonuclease (APE) DNA repair proteins APE1 and APE2 in GC B cells, where APE1 expression is surprisingly low and the homolog APE2 is highly expressed. In the absence of APE2, SHM is reduced ∼2-fold in Peyer’s patch (PP) GC B cells, suggesting that repair of AID lesions by APE2 may be error-prone (1), particularly when APE1 expression is limiting. However, APE2-deficient GC B cells also have reduced proliferation and/or survival (2) that could impact mutation frequency. In this study, we directly test the hypothesis that APE2 promotes and APE1 suppresses SHM.

Both CSR and SHM are initiated by AID, which converts cytosine to uracil in DNA (dU) (3, 4). Whereas the removal of uracil from DNA is normally accomplished by the highly accurate and efficient base excision repair (BER) and mismatch repair (MMR) pathways, dU repair is error-prone at the Ig locus in proliferating B lymphocytes, resulting in mutations and DNA double-strand breaks (DSBs) (5). The DSBs are necessary for CSR, and the mutations promote increased affinity of Ab when coupled with selection mechanisms in GCs (3, 4, 6, 7). Both processes are essential for maturation of the immune response but are also associated with genomic instability such as mutations, translocations, and tumorigenesis, including diffuse large B cell lymphomas that are GC cell derived (8).

During accurate repair in nonmutating cells, dU is excised by uracil DNA glycosylase (UNG), leaving abasic (AP) sites that are recognized by APE. APE makes a single-strand break (SSB) with a 3′ OH that is usually extended by DNA polymerase (Pol) β (POLB) to replace the excised nucleotide (9). During SHM, dUs may be replicated over before they can be excised and are read as dT by replicating DNA polymerases, resulting in dC-to-dT transition mutations (3). Also, some AP sites go unrepaired and encounter replication, leading to the nontemplated addition of any base opposite the AP site. However, it is not clear why dUs and AP sites escape accurate repair by the highly efficient enzymes UNG and APE1 and lead instead to mutations.

The MMR pathway competes with UNG for repair of U:G mismatches. MMR excises a patch surrounding the mismatch, and resynthesis during SHM involves error-prone translesion polymerases (TLPs) such as Pol η, Pol ζ, and Rev1. The MMR pathway is required for the generation of A:T mutations during SHM, and it also generates DBSs in switch regions that are necessary for CSR (10–13). The collision of BER and MMR pathways, both of which can repair U:G mismatches, and the use of a “noncanonical” MMR pathway during the G₁ phase (14), when AID is active (15), partially explains error-prone repair of AID-induced lesions, contributing to mutations and DNA break formation. However, both MMR and BER are active in cultured B cells that undergo CSR and hypermutation at switch regions (11) where SHM of the V genes remains undetectable (16–19).

There are two APE homologs, which we found are differentially expressed in the GC relative to cultured B cells (1). APE1 is the major mammalian AP endonuclease. It is ubiquitously expressed, highly efficient, and essential for early embryonic development in mice and for viability of human cell lines (20–22). APE1 interacts with XRCC1, which coordinates SSB repair via POLB and ligase (23, 24). In contrast, APE2 is a nonessential homolog with very weak AP endonuclease activity (25, 26). APE2 lacks the N-terminal XRCC1-interacting domain of APE1, instead having a unique C terminus with a proliferating cell nuclear Ag (PCNA)–interacting domain (27–29). PCNA is known to interact with several proteins that are necessary for SHM, including UNG, MMR proteins, and error-prone TLPs (30–34). Expression of APE1 is very low in GCs, where APE2 expression is highly induced (1, 35, 36). In APE2-deficient GC B cells, both SHM frequency and the spreading of mutations beyond AID hotspots is reduced (1, 37), suggesting that repair pathways using APE2 might be error-prone. In contrast to the GC, activated, cultured B cells express APE1 and APE2 more equally, and SHM of the V genes does not occur.

In this study, we investigated APE1 and APE2 expression during B cell activation and asked how perturbation of APE1 and APE2 levels impacts CSR and SHM. To avoid the complication of reduced proliferation and/or survival of APE2-deficient GC B cells, we developed an in vitro system where APE2-deficient B cells proliferate normally (35), and where we can manipulate the ratio of APE1/APE2 expression levels. We found that APE2 is indeed error-prone, promoting CSR and mutations in the Sμ region. Furthermore, APE2 also promotes AID-dependent SHM in the V region J_H4 intron in cultured B cells, but only when APE1 levels are reduced. Although APE1 promotes CSR, it suppresses SHM.

All mouse strains were backcrossed to C57BL/6 for more than eight generations (except B1-8, five generations) and prior to interbreeding to create double-deficient mice. AID-deficient mice were obtained from T. Honjo (Kyoto University, Kyoto, Japan). Apex1^+/− mice were obtained from E. Friedberg (38) (University of Texas Southwestern Medical Center, Dallas, TX). MSH2-deficient mice were obtained from T. Mak (University of Toronto, Toronto, ON, Canada), and UNG-deficient mice were obtained from T. Lindahl and D. Barnes (London Research Institute, London, U.K.). APE2-deficient mice (39) and B1-8 mice (40) were previously described. Because apex2 is on the X chromosome, we used male apex2^Y/− mice in all experiments. The wild-type (WT) mice were littermates of either the apex2^Y/− or apex1^+/−apex2^Y/− mice. Mice were housed in the Institutional Animal Care and Use Committee–approved specific pathogen-free facility at the University of Massachusetts Medical School. The mice were bred and used according to the guidelines from the University of Massachusetts Institutional Animal Care and Use Committee.

B cells were enriched from mechanically dispersed spleen by T cell depletion with Ab and complement followed by centrifugation over Lympholyte (Cedarlane), as previously described (19), and activated with LPS, anti–IgD-dextran, and BLyS-FLAG. IFN-γ was added to cultures for IgG2a CSR. Cells were infected with retrovirus (RV), described below, 24 h after activation and cultured for 2 or 4 additional days for CSR and SHM analysis, respectively. Tamoxifen was added to cultures with pMX-PIE RV to force the estrogen receptor (ER)-tagged construct into the nucleus. For GLSμ and J_H4 mutation analysis, four independent cultures for each RV infection were set up to maximize the potential for unique clones and unique mutations. At 2 d postinfection, cultures were split and fed with fresh medium, tamoxifen, and switch inducers to promote viability and continued cell division. Viable cells were isolated by flotation on Lympholyte at 4 d postinfection, and GFP⁺IgM⁻ and GFP⁺IgM⁺ cells from each culture were sorted independently on a FACSAria IIu (BD Biosciences) after staining with anti–IgM-PE (SouthernBiotech) and 7-aminoactinomycin D. Two or three experiments were performed for each RV analyzed. There was no difference in viability in any of these cultures (Supplemental Fig. 2). SHM was analyzed in IgM⁻ cells, which have undergone CSR and therefore experienced AID activity, and GLSμ was analyzed in IgM⁺ cells, which have the μ switch region intact.

pMX-PIE-AID-FLAG-ER-IRES-GFP-puro (41) was received from Drs. V. Barretto and M. Nussenzweig (The Rockefeller University, New York, NY). The control RV, pMX-PIE-ER-IRES-GFP, was constructed and viruses were prepared as previously described (42). pMIG (43) was received from Dr. J. Chaudhuri (Sloan-Kettering Memorial Cancer Center, New York, NY). Full-length APE1, APE2, and truncation mutants were generated by PCR, cloned into Bluescript (Stratagene), confirmed by sequencing, and subcloned into pMX-PIE and pMIG. To create the APE2 PIP2 substitution mutant, the APE2 gene in Bluescript was mutated using QuikChange (Stratagene), sequenced, and then reinserted into pMX-PIE and pMIG. Retroviruses were produced in Phoenix-E cells with pCL-Eco and X-tremeGENE (Roche).

Pelleted cells from FACS-purified or cultured B cells were lysed in RIPA buffer, and 15–20 μg of whole-cell extracts was analyzed on 8% polyacrylamide gels as described previously (19) with polyclonal goat anti-APE1 (R&D Systems), rabbit anti-APE2 (AnaSpec) (35), anti-POLB, rabbit anti-Grb2 (growth factor receptor–bound protein 2), anti-ER, and anti-GAPDH (Santa Cruz) primary Abs, with goat anti-rabbit and donkey anti-goat secondary Abs coupled to HRP (Santa Cruz).

DNA was prepared from FACS-purified cells by proteinase K and RNase A digestion and ethanol precipitation. To assay SHM, a 492-bp fragment of the V_HJ558L framework 3-J_H4-3′ flanking region was amplified by a nested PCR using (Stratagene); primers were modified slightly from those in McDonald et al. (44). Primers for the first amplification were forward, 5′-AGCCTGACATCTGAGGAC-3′ and reverse, 5′-GTGTTCCTTTGAAAGCTGGAC-3′. Nested primers for the second amplification were forward, 5′-CCGGAATTCCTGACATCTGAGGACTCTGC-3′ and reverse, 5′-GATGCCTTTCTCCCTTGACTC-3′. The reaction conditions for the first primer set were 95°C for 30 s, 57° for 30 s, and 72° for 1 min for 30 cycles; and for the second primer set were 95° for 30 s, 57° for 30 s, and 72° for 1 min for 35 cycles. PfuTurbo (Agilent) (error rate = 1.3 × 10⁻⁶) was used to amplify a 749-bp 5′ GL Sμ fragment as described (45). Primers were 5u3 (forward primer), 5′-AATGGATACCTCAGTGGTTTTTAATGGTGGGTTTA-3′ and m2R (reverse primer), 5′-GCTACTCCAGAGTATCTCATTTCAGATC-3′ (46).

The PCR products were electrophoresed on 1% agarose gels, the Sμ and J_H4 bands were purified using a QIAquick gel extraction kit (Qiagen), and dA tails were added with Taq polymerase, cloned using a TOPO TA cloning kit (Invitrogen), and sequenced by Macrogen (Cambridge, MA). We selected only 24 colonies from each of the four independent cultures for a total of 96 clones sequenced from each experiment to avoid repeat CDR3s. Unmutated sequences were not excluded from mutation frequency calculations.

PPs were excised from 12- to 20-wk-old, naive, WT mice, mechanically dispersed on ice, and passed through 45-μm nylon mesh. For intracellular APE1 analysis, PP or cultured B cells were stained with Live/Dead Blue (Thermo Fisher Scientific) (1:200) in PBS, washed and surface stained with B220 VioGreen (1:50), GL7 FITC (1:300), CD95 PE (1:50), CD86 PE-Cy7 (1:100), and CXCR4 Per-CP-EF710 (1:50), then washed with PBS, fixed with 2% formaldehyde, and permeabilized with 0.1% saponin buffer prior to staining with goat IgG anti-APE1 (R&D Systems) (1:120) in saponin buffer, and then donkey anti-goat DyLight 649 (1:300, Jackson ImmunoResearch) in saponin buffer, washing with saponin buffer after each step. Fc Block was added prior to staining. For proliferation studies, cells were stained with CellTrace Violet (Thermo Fisher Scientific) prior to culture, and B220 FITC (1:200) prior to APE1 intracellular staining as above. Analysis was done on a four-laser (R/B/V/YG) Aurora cytometer (Cytek Biosciences). The following Abs were used for FACS purification and genomic analysis: REAffinity mAbs CD45R/B220 (REA755, VioGreen) and CD95 (REA453, PE-Vio770) (Miltenyi Biotec) and biotinylated GL7 (IgM, BioLegend), followed by SA-allophycocyanin/Cy7 (BioLegend) and a Live-or-Dye NucFix fixable dead cell staining kit (Biotium).

Genomic profiling was achieved using templated oligonucleotide sequencing (TempO-Seq, BioSpyder Technologies). Briefly, integrated (optional) intracellular staining with mRNA in situ hybridization (icTempO-Seq) was carried out on nonsorted cell suspensions after surface staining (optional), dead cell labeling, fixation, and permeabilization. Cells (∼10⁵/sample, in multiples when necessary) were then intracellularly stained using a mouse whole-transcriptome array with DO (DNA oligonucleotide) probe sets (msWT-Assay; 30,146 dual DO sets covering 21,450 distinct mouse gene transcripts) at 0.2 nM in 100 μl per sample of hybridization solution (BioSpyder Technologies), supplemented with 0.1% saponin and murine RNase inhibitor (M0314, New England Biolabs), then overlayered with mineral oil and incubated at 45°C overnight. The next day, cells were washed and DO probes were ligated in situ with ligation buffer (BioSpyder Technologies) for 1 h in a H₂O-saturated 37°C, CO₂ incubator. After another wash, cells were suspended in FACS sample buffer and sorted on a BD FACSAria II using stringent doublet- and dead cell–exclusion gating. Sorted populations typically in a four-way sort setup included the following: single GC B cells (B220⁺, CD95⁺, GL7⁺), follicular naive B cells (B220⁺, CD95⁻, GL7⁻), and in vitro–cultured B cells (B220⁺, CD95⁺, GL7⁺) using the same staining and hybridization protocol. All sorted subsets were >95% pure upon reanalysis. One hundred cells/sample were accurately adjusted either by plate sorting, in which case a second round of FACS sorting with identical gates was used, or by precisely adjusting FACS-sorted cell concentrations by aliquot counting (MACSQuant analyzer using the volume counting option).

Sequencing libraries were prepared from FACS-purified cells (100 cells/sample) by one 28-cycle and single-plex barcoding PCR (index PCR plates and PCR settings, BioSpyder Technologies), followed by library pooling and clean-up (NucleoSpin, Macherey-Nagel). Libraries were run on NextSeq high output (∼450 million reads) or HiSeq (∼250 million reads/lane) with pooled samples requiring ∼3 million reads/sample of WT array. Sequences were aligned and statistically analyzed using the TempO-Seq-R cloud-based software package (BioSpyder Technologies), applying the DeSeq2 method for differential analysis of RNA sequencing data (47).

APE1 expression previously assessed by Western blotting appears very low in the GC relative to activated, cultured B cells, whereas APE2 is highly expressed in both (1). To better understand the dynamics of APE1 expression, we sought to determine the distribution of expression at the single-cell level in both GC and cultured B cells. Because Western blotting provides only an average expression level, we developed an intracellular stain to track APE1 by flow cytometry. We first asked whether expression differed between the dark zone (DZ), where rapid proliferation and mutation take place, and the light zone (LZ), where selection occurs and B cells are in transit, entering and exiting the GC. We found that, compared with activated cultured B cells that express nearly uniform, high levels of APE1 (Fig. 1A, top row; 83% APE1^hi), GC B cells express varying levels of APE1 and only ∼20% are APE1^hi (Fig. 1A, middle row). APE1^lo GC cells comprise both LZ and DZ B cells, whereas APE1^hi cells are skewed toward the LZ (Fig. 1a, bottom row, 1B). Interestingly, APE1^hi cells are enriched in CXCR4/CD86 double-positive cells, which we speculate might be in transition (Fig. 1a, bottom right). We conclude that low APE1 expression in the GC correlates with the proliferation and mutation that occur in the DZ.

We then evaluated regulation of APE1 expression during B cell activation in culture where CSR but not SHM occurs. We found that APE1 RNA and protein levels are both high up to 48 h postactivation and then begin to decline (Fig. 1C, 1D). In contrast, APE2 RNA levels are steady throughout the culture. Using CellTrace Violet to track cell division, we found that APE1 protein levels are stable and independent of cell division at 2 d (48–53 h) postactivation when RNA is high. However, in cells harvested after 2.5 d (∼67 h) in culture, APE1 decreases steadily in proportion to the number of cell divisions (Fig. 1E–G), even though cells were restimulated every 2 d to promote proliferation. Thus, there is a time-dependent decrease in expression, consistent with the decrease we see in mRNA expression, but in addition, the remaining APE1 protein is diluted out by cell division. Biologically, high APE1 expression correlates with CSR activity, which peaks 2–3 d after activation, and low APE1 occurs after cells have undergone many rounds of cell division, as in the GC, correlating with SHM. Indeed, after the initial activation phase (24–48 h), the more times a cell divides, the less APE1 it has.

We next sought to determine the impact of the dynamic changes in APE expression on SHM and CSR. We developed a culture system where we could manipulate the expression levels of APE1 and APE2 in ex vivo primary spleen B cells both genetically and by RV overexpression. Unlike GC B cells, these cells historically do not undergo SHM but do proliferate normally in the absence of APE2, presumably due to higher APE1 expression. apex2 is on the X chromosome, so APE2-deficient cells from male apex2^Y/- mice were used. Because APE1 deficiency is embryonic lethal, we used apex1^+/− mice that are haploinsufficient to reduce APE1 expression (38). To reduce APE as much as possible, we crossed apex1^+/− and apex2^+/− mice. These mice are haploinsufficient for APE1 and APE2-deficient, which we term DBL (double: apex1^+/−apex2^Y/−). Splenic B cells were activated with LPS, anti–IgD-dextran, IFN-γ, and BLyS, and then infected with RV to express ER-tagged APE1, APE2, or ER tag alone (Fig. 2), or untagged proteins (Supplemental Fig. 1A, 1B), all expressed with a bicistronic GFP reporter. Tamoxifen was added during retroviral infection to force the ER-tagged construct into the nucleus. Expression levels of endogenous APE1 and APE2, as well as RV APE2-ER, are shown in these cultures by Western blotting (Fig. 2A).

We first evaluated the effect of perturbed APE1 and APE2 levels on CSR, because we found previously that both APE1 and APE2 contribute to DSB formation in Sμ and to optimal CSR (35). We find in the present study that neither APE1 nor APE2 is limiting for CSR in WT B cells, as overexpression of neither RV-APE1-ER nor RV-APE2-ER altered CSR to IgG2a (Fig. 2B). As in our previous report, CSR in APE2-deficient B cells (apex2^Y/−/ER) is reduced to 60% of WT levels (WT/ER) and we now find that CSR is fully restored by expression of RV-APE2 (apex2^Y/−/APE2, Fig. 2C). In DBL B cells, CSR is further reduced to only 40% of WT levels (Fig. 2D), where expression of RV-APE2 restores CSR to ∼70% of WT CSR. Overexpression of RV-APE1 does not increase CSR in B cells that are deficient for APE2 (apex2^Y/− or DBL). The same results were obtained using untagged APE proteins and for CSR to IgG3 (Supplemental Fig. 1A, 1B). These data support our previous conclusion (35) that APE1 and APE2 work together for optimal CSR.

We then asked whether APE1 and APE2 levels impact the AID-dependent mutations that accumulate in GLSμ prior to successful switch recombination. B cells were activated for CSR and infected with RV, but then fed with fresh activators and tamoxifen on day 3 and cultured for an additional 2 d prior to FACS purification of RV-infected (GFP⁺) cells that remained IgM⁺, indicating lack of successful CSR. These cells have a high mutation rate at GLSμ (WT/ER, 9.5 × 10⁻⁴ mutations/bp) that is AID-dependent. The mutation frequency and number of mutations per sequence are shown in Fig. 2E, Supplemental Fig. 2, and Supplemental Table I. B cells with reduced APE1 (apex1^+/−/ER) have a 1.3-fold, highly significant increase in mutations versus WT/ER, consistent with an accurate repair role for APE1. However, the opposite is observed for APE2, where mutations are reduced by almost half in the absence of APE2 (apex2^Y/−/ER versus WT/ER) and are restored by expression of RV-APE2 (WT/ER versus apex2 ^Y/−/APE2-ER, NS). In DBL B cells, mutations are also reduced by half, and again restored by RV-APE2. Note that the increase in mutations seen with reduced APE1 (apex1^+/−/ER) occurs only when APE2 is present, that is, not in DBL B cells. Surprisingly, when APE1 is overexpressed, the GLSμ mutation frequency increases slightly relative to control (DBL/APE1-ER versus DBL/ER, p = 0.02). However, the increased frequency is due to one highly mutated sequence (Supplemental Fig. 2, DBL/APE1) and includes a high frequency of insertions and deletions (indels), which can be indicative of DNA breaks (Fig. 2F), consistent with the highly efficient endonuclease activity of APE1. Interestingly, even though there is a 4-fold increase in indels with DBL/APE1-ER expression, this condition, which lacks APE2, does not promote efficient CSR (Fig. 2D). Indels are also increased in apex1/ER cells, but these cells express endogenous APE2, and CSR is only modestly reduced in apex1^+/− B cells (≈80% of WT) (35). As such, when levels of APE1 and APE2 are perturbed, APE2 expression and the frequency of GLSμ mutations correlate better with efficient CSR than do APE1 expression and indel mutations, suggesting that the indels may indicate breaks that are processed in a way that does not promote CSR. We conclude from these data that APE1 and APE2 work together to promote efficient CSR and that low-level APE1 expression is sufficient. Furthermore, APE2 contributes significantly to the generation of AID-dependent mutations at GLSμ.

Our GLSμ analysis suggests that AP site repair by APE2 may be error-prone, in agreement with our findings in PP GCs, where both very low APE1 and high APE2 expression contribute to SHM of V genes (1). In contrast, cultured B cells highly express both APE1 and APE2 (1, 35), and SHM of the V genes does not occur. To test our hypothesis that APE2 actively promotes mutations, we used the same culture system as for GLSμ mutations and examined the impact of altered APE1/APE2 expression levels on V region mutation frequency.

In apex1^+/− B cells expressing RV-APE2, the ratio of APE2/APE1 was determined by Western blotting as a function of time in culture (Fig. 3A), and after 48 h it is comparable to what we previously observed in GC B cells (1). The arrows at the bottom of Fig. 3A indicate when fresh stimulators and/or RVs are added. RV-APE2 is detectable at constitutively high levels throughout days 2–5 of the culture at much higher levels than endogenous APE2, which is induced upon activation and then decreases, cycling with repeated stimulations on days 0, 1, and 3. apex1^+/− B cells express about half as much APE1 as do WT cells, and levels decrease slowly and steadily with time in culture, despite repeated stimulation and in agreement with our intracellular stain results for APE1 (see Fig. 1). Interestingly, endogenous POLB also appears to cycle with activation as does APE2. This culture system mimics GC B cells with rapid proliferation, multiple rounds of stimulation, low levels of APE1, and high levels of APE2 expression.

To analyze SHM in vitro, we used DBL B cells so that APE1 is low, and APE2 is deleted and can be replaced with WT or mutated APE2. On day 5 of culture, we purified GFP⁺IgM⁻ cells, which had undergone CSR and therefore experienced AID activity, and cloned and sequenced the J_H4 intronic region that lies 3′ to recombined J558VDJ_H4 genes, the most abundant V gene family in mice. The J_H4 intronic region was analyzed for SHM in GC cells as it undergoes mutation but is not subject to selection. As previously shown by others, we found no significant difference in mutations between WT and AID-deficient cultured B cells. However, in APE1 haploinsufficient B cells (apex1^+/−/ER) there is a significant, 2-fold increase in mutations compared with AID-deficient B cells (aid^−/−/ER), similar to our results at GLSμ. With added overexpression of RV-APE2 (apex1^+/−/APE2-ER), we see a further increase in mutations that is 2.7-fold higher than aid/ER and 1.9-fold higher than WT/ER (Fig. 3B, Supplemental Table II). APE2-deficient cells with WT APE1 levels (apex2^Y/−/ER) also have increased mutations (including a rare-occurring indel) and re-expressing APE2 in these cells (apex2^Y/−/APE2-ER) brings mutations back to WT levels. This suggests that it is not merely APE2 or APE2-ER expression, but rather perturbation of APE1/APE2 levels that promotes SHM. However, decreased APE1 levels do not promote SHM in the absence of APE2 (DBL/ER), whereas mutations in DBL B cells expressing APE2 (ER or untagged) are increased 2.4- and 1.8-fold, respectively, relative to WT/ER control. The slightly higher mutation rate in DBL/APE2-ER is not significantly different from DBL/APE2-pMIG or apex1^+/−/APE2-ER. Importantly, overexpression of APE2 does not induce J_H4 mutations in cultured B cells when APE1 is at normal levels (apex2^Y/−/APE2-ER). The mutation frequency in DBL/ER and WT/ER control cultures was similar to aid^−/−/ER B cells, and to that reported for J_H4 mutations in B1-8 aid^−/− cultured B cells (0.9 × 10⁻⁴) (18). Sequences with identical CDR3s are rare (∼5%) and were omitted from analysis. The unique CDR3s indicate that we are not merely expanding clones with prior mutations. To show that perturbed APE levels do not cause random genome-wide mutations, we sequenced a 595-nt segment of msh6 DNA from DBL/APE2-ER cells, and the mutation frequency was similar to the aid^−/−/ER background (Supplemental Table II).

To further validate that mutations were AID-dependent, we tested msh2^−/−ung^−/− B cells infected with control RV-ER, where AID-induced dUs will go largely unrepaired and read as C→T or G→A transition mutations. The mutation rate in these cells was 4.9 × 10⁻⁴, which is 5-fold higher than aid^−/−/ER background and 2-fold higher than in DBL/APE2-ER cultures (Supplemental Table II). Furthermore, 17 out of 18 mutations were G:C-to-A:T transitions, as expected when mutations are induced by AID. Also, APE2-ER did not promote mutations in DBL B cells in the absence of AID (Supplemental Table II; see DBL × aid^−/−/APE2-ER). These data show that AID does act on V(D)J genes in cultured B cells, but that the dUs are normally repaired by error-free mechanisms, which we show in the present study to include APE1. Thus, engineering APE levels to resemble those seen in the GC promotes low-level AID-dependent SHM at the J_H4 intron in cultured primary B cells.

Because the mutation frequency at the J_H4 intron in culture is much lower than occurs in the GC, we asked whether SHM would be higher in the VDJ exon by using B cells from B1-8 mice (40). We backcrossed B1-8 mice to the C57BL/6 background for five generations and then bred B1-8 mice that were MSH2- and UNG-deficient, or DBL, with or without AID deficiency. We saw a 3-fold increase in mutations in B1-8/msh2^−/−ung^−/− mice versus B1-8/aid^−/− mice, all of which were transitions at G:C bp, again indicating AID activity at the V region (Supplemental Table II). However, this was unexpectedly lower than the mutation frequency we detect at J_H4 in C57BL/6, and, surprisingly, APE2 overexpression had no effect on SHM in DBL B1-8 B cells (Supplemental Fig. 3C). Because we showed above that APE2 does not increase mutation frequency unless APE1 is decreased, we examined APE1 expression levels in B1-8 B cells and found that APE1 levels are indeed higher in B1-8 cultured B cells relative to C57BL/6 B cells by Western blotting and by flow cytometry (Supplemental Fig. 3A, 3B, 3D). Increased APE1 expression is also seen in vivo in this widely used mouse model, as shown in PP GC cells (Supplemental Fig. 3E, 3F). Although we did not observe increased mutation frequency in the VDJ exon as expected, our results in the B1-8 model further demonstrate that APE2 does not increase SHM unless APE1 expression is low.

To try to understand how APE2 contributes to SHM, we explored the domains of APE2 that could regulate accuracy of repair. APE1 and APE2 share a core domain that is highly conserved, with the exception of 2 aa changes in the active site that reduce the relative endonuclease activity of APE2 by several orders of magnitude (26). Also, as shown in Fig. 4A, APE1 has a unique N terminus that interacts with XRCC1 (23, 24), whereas APE2 has a unique C terminus with three domains that are conserved between mice and humans: 1) aa 314–364, of unknown function; 2) a functional PCNA-interacting motif (PIP-box; aa 390–397); and 3) a zinc finger (Zf-GRF) domain (aa 455–516) that interacts with ssDNA and activates ATR/Chk1 to stimulate SSB repair (27–29). To explore whether these domains impact the role of APE2 in CSR and SHM, we made a series of truncation mutants and also mutated 2 aa in the PIP domain that were previously shown to block PCNA interaction with human APE2 (28) (Fig. 4A). All are stably expressed in B cells (Fig. 4B) and do not impact viability (Supplemental Fig. 4). We found that APE2-ΔC, which lacks the entire unique C terminus, cannot restore CSR in DBL B cells (Fig. 4C). APE2-ΔD, which lacks the PIP and Zf-GRF domains, partially restored CSR (1.4-fold over RV-ER control, p < 0.005), whereas APE2-ΔE, which has the PIP domain but lacks the Zf-GRF domain, fully restores CSR, frequently to a higher level even than full-length APE2. The same results were found in cells expressing the untagged proteins, although APE2-ΔD was even more effective without the C-terminal ER tag and not significantly different from full-length APE2 (Supplemental Fig. 1C). APE2-PIP2, with two key amino acids in the PIP domain mutated, was also able to fully restore CSR (Fig. 4C). Although mutations in the PIP domain did not reduce CSR efficiency, we could not rule out a role for PCNA, because PCNA has been shown to interact functionally with APE2 via the Zf-GRF domain in addition to the PIP box (29). As this domain is intact in APE2-PIP2, this mutant might still interact with PCNA. We conclude that part of the C terminus is required for the role of APE2 in CSR, but the Zf-GRF domain is not.

GLSμ mutations were lower in all of the mutants, although APE2-ΔE–expressing cells had mutation frequencies near those of full-length APE2 (Fig. 4D). Interestingly, APE2-ΔE caused a highly significant increase in indels (Fig. 4E), similar to that seen in APE1 overexpressing cells (see Fig. 2F). This suggests that the C-terminal ATR-ChK1–interacting domain that is lacking in APE2-ΔE might indeed stimulate SSB repair, thus slightly suppressing CSR. As DBL/PIP2 cells also had increased indels, we speculate that this C-terminal mutation might also interfere with ATR–ChK1 interaction.

J_H4 mutations are differently affected by the APE2 mutants than are CSR and GLSμ mutations. Expression of all APE2 mutants increased J_H4 mutations relative to ER control, although APE2-ΔC was of borderline significance (p = 0.06; Fig. 4F). APE2-ΔD–expressing cells have the highest frequency of mutations. APE2-ΔD appears to be the most highly expressed by Western blot in Fig. 4B; however, this could reflect transfection efficiency because mutation and CSR analyses were gated for GFP⁺ cells, whereas the extracts for Western blot analysis were made without sorting for GFP⁺ cells expressing RV. RV-APE1 did not increase J_H4 mutation frequency. We conclude that, even without the C terminus, the inefficient endonuclease activity of APE2 promotes mutations in these cultures.

Analysis of the J_H4 mutations is shown in Supplemental Table II, although the low number of mutations relative to background, random mutations makes it difficult to draw conclusions. However, one observation is noteworthy; that is, indels were detected in J_H4 only in three conditions—APE2 deficiency (apex2^Y/−/ER) or RV-APE1 or APE2-ΔE overexpression (Supplemental Table II)—similar to what we found in GLSμ. This is consistent with a high APE1/APE2 ratio favoring DNA breaks, and also with the APE2 C-terminal ATR/Chk1 domain that is absent in ΔE stimulating SSB repair in culture. The presence of this domain in WT APE2 presents another example of the balance between accurate and error-prone repair that occurs during CSR and SHM.

Because the pattern of APE1/APE2 expression differs between GC and cultured B cells and impacts mutation frequency, we performed whole-transcriptome profiling on both populations to gain further insight into the gene expression context of in vitro–stimulated B cells exhibiting a low mutation frequency. Genomic profiling was achieved using TempO-Seq (BioSpyder Technologies) (48), a modified version of RNA annealing, selection, and ligation with massive parallel sequencing (RASL-seq) adapted for formalin-fixed, paraffin-embedded samples (49–51), and further adapted to permit intracellular staining prior to FACS purification of B cell subsets. Of 21,450 genes surveyed, 11,780 were detected, of which 1,711 (14.5%) were significantly downregulated and 3,219 (27.3%) were upregulated (adjusted p < 0.05) in cultured versus GC B cells. The expression profile of MMR and BER pathway genes, as well as genes selected for their association with CSR, SHM, and/or GCs, is shown in Fig. 5. Many genes associated with SHM (aicda, bcl6, msh6, rev1, polh, apex2) are expressed at higher levels in GC B cells, as are several DNA repair genes whose role is less clear (neil1, neil3, mbd4, mdm2). Several genes known to promote CSR are expressed at higher levels in cultured B cells [apex1, the de-ubiquitinase usp22 (52), and the transcription factor batf (53)]. The increased ratio of APE2/APE1 expression in GC cells is seen in this study as a combination of increased APE2 and decreased APE1 relative to cultured B cells. The entire dataset can be accessed at https://doi.org/10.5281/zenodo.7617207.

In conclusion, the results support and extend our previous finding that APE1 and APE2 work together for optimal CSR (Figs. 6, 7) (35). Furthermore, when APE1 is limiting, as in the GC, repair by APE2 becomes error-prone, likely due to its inefficient AP endonuclease activity and possible interaction with PCNA and TLPs. APE2 promotes V region SHM and GLSμ mutations, whereas APE1 promotes DNA breaks but suppresses SHM in vitro. The APE2 C terminus is required for its role in CSR but not SHM, although the impact of its interaction with PCNA merits further study.

Historically, SHM of the J_H4 intronic region has been undetectable in B cells activated in vitro (16–19), but it has never been examined in cells with reduced APE1. We genetically modified splenic B cells to mimic the low APE1 and high APE2 expression seen in the GC and can now detect AID-dependent SHM in primary cultured B cells. Our findings demonstrate the importance of the ratio of APE1/APE2 expression levels for both SHM and CSR. Both processes have long been associated with the GC; however, it was recently shown that CSR activity declines in the GC (36), where APE1 expression is very low. APE1 creates DSBs required for CSR, but DSBs are not an intermediate during SHM. Indels that can result from DNA breaks are abundant in the noncoding switch regions in cells undergoing CSR, whereas SHM is limited primarily to point mutations, where indels in the Ig coding region would be deleterious to function. APE1 is not required for SHM in cell line models (54), and haploinsufficiency has no effect on SHM in vivo (1), although it does reduce CSR (35, 55). In the GC, low APE1 expression would limit the deleterious effects of DNA breaks, whereas increased APE2 expression promotes SHM (1) and supports B cell viability (2).

APE2 seems uniquely suited for error-prone repair. Its catalytic site is more open, able to accommodate insertion of a mispaired base (26). Also, APE2’s inefficient endonuclease activity, relative to APE1, would leave more abasic sites unrepaired, causing transition and transversion mutations at the G:C bp attacked by AID. Furthermore, APE2’s C terminus can interact functionally with PCNA (27, 28), which increases its exonuclease activity, while also interacting with error-prone TLPs and MMR (30–34) that promote SHM. In the presence of PCNA, the exonuclease activity of both APE2 and MMR-associated EXO1 are more processive, so a larger gap can be excised and filled in by error-prone TLPs such as Pol η, Pol ζ, and REV1.

Based on our findings in the present study, we propose a model incorporating how changes in APE1 and APE2 expression during B cell activation might shift the balance from accurate to error-prone repair, and how APE1 and APE2 might work together to promote CSR. First, in a resting B cell, APE1 predominates, and its interaction with XRCC1 promotes accurate repair by POLB and ligase (Fig. 6A). Then, during early activation (pre-GC) when both are expressed, APE1 and APE2 work together to turn multiple AP sites into DSBs for CSR. When APE1 and APE2 cleave two AP sites on opposite strands in very close proximity, CSR independent of MMR can be explained by the 3′ to 5′ exonuclease activity of APE2. When stimulated by PCNA, APE2 can excise ∼10–12 nt (28), and could encounter a nick made by APE1 on the opposite strand, forming a DSB (Fig. 6B). When AID lesions are farther apart, MMR is required (11). In this scenario, APE2 interacts with MMR via PCNA, and 5′ to 3′ strand excision by MMR-EXO1 results in DSB formation when a nick on the opposite strand is encountered (Fig. 6C). CSR in Msh2-deficient cells is reduced by 2-fold, as are GLSμ DSBs (10, 11, 56). These two models are similar in requiring exonuclease activity and are not exclusive. The distance and strand orientation between AP sites could dictate which exonuclease is used to create a DSB, as PCNA stimulates the processivity of both MMR-EXO1 and APE2 exonuclease. We tested the idea of partial redundancy between APE2 and MMR by generating DBL mice that are also deficient in MSH2. Indeed, the decrease in CSR in MSH2- and APE2-deficient B cells is additive, with CSR reduced between 3- and 6.5-fold in DBL/MSH2-deficient cells (Supplemental Fig. 5). Although APE1 likely makes more SSBs due to its efficient endonuclease activity and ability to interact with AID (57), the nicks made by APE2 might be more likely to undergo excision, contributing to DSB formation. Recent reports describe direct interaction between APE1 and APE2 during SSB repair, and a role for APE2 in SSB end resection (29, 58–60).

Finally, with rapid cell division, APE1 expression is diminished and APE2 predominates in the GC (Fig. 6D). In our model, a complex containing UNG, APE2, and PCNA recruits TLPs and MMR during the G₁ phase when AID is active (11, 15, 61, 62) and when noncanonical MMR can be error-prone (63). Similar to APE2, UNG also has a canonical PCNA-interacting domain. APE2 may also be responsible for making the nicks in DNA that permit entry of MMR and EXO1. Consistent with this, there is a 2-fold reduction in SHM in PP GC cells and a significant reduction in spreading of mutations beyond AID hotspots in apex2^Y/− mice (1). The decrease is not dramatic, but redundancy in the ability of APE2, UNG, and MMR to all interact with PCNA likely reduces the impact of APE2 deficiency. Indeed, there is a surprising, additive effect of combined UNG and APE2 deficiency on SHM in the GC, where A:T mutations are further decreased compared with the single mutants (1). This was not expected because UNG acts upstream in the same pathway as APE2, but it is consistent with some redundant function such as PCNA recruitment.

Although we did not directly demonstrate APE2–PCNA–MMR interaction in our system, we base our models on previously published functional interactions of these proteins. In addition to PCNA, the C-terminal Zf-GRF domain of APE2 has been shown to interact functionally with ATR and Chk1 to promote SSB repair (29). This function of APE2 is essential for survival of BRCA2-deficient cells (64) but is unlikely to be beneficial to CSR or SHM. Disruption of ATR/Chk1 activation would result in delayed SSB repair, which would be beneficial to both CSR and SHM, and these proteins are suppressed by BCL6 in the GC (65, 66). Indeed, CSR was slightly, although not significantly, increased in B cells expressing APE2-ΔE (Fig. 4C, Supplemental Fig. 1C), which lacks the Zf-GRF domain associated with ATR/Chk1 activation.

We also modeled how SSB repair in the GC is repressed through a multifaceted approach (Fig. 7), where 1) low APE1 expression in the GC reduces XRCC1/POLB/ligase-mediated accurate repair of AP sites made by AID and UNG, and 2) APE2 expression is increased, but its potential to interact with ATR and Chk1 is inhibited because these proteins are suppressed by BCL6. This leaves APE2 able to promote error-prone repair through 1) its inefficient AP endonuclease activity, 2) its interaction with PCNA and associated TLPs, and 3) its enhanced exonuclease activity stimulated by PCNA. Furthermore, with APE1 levels limiting in the GC, survival of these rapidly proliferating B cells is highly dependent on APE2 for the repair of both AID-dependent damage and AID-independent oxidative damage (2).

In contrast, cultured B cells express APE1 and proliferate normally in the absence of APE2 (35). APE2 can promote mutations in vitro when APE1 expression is reduced, but the mutation frequency we observe at J_H4 in culture is very low compared with that of GC B cells; too low to permit analysis of the mutation spectrum. A likely reason for the low mutation frequency in culture is that we cannot eliminate APE1, and residual APE1 levels in DBL (apex1^+/−) cells are sufficient to compete with APE2 to promote accurate repair of some AID lesions. Furthermore, recent studies clearly showed that AID associates poorly with V genes in cultured B cells and B cell lines relative to GC B cells, and relative to its association with Sμ (17, 18). Maul et al. (18) showed that the accumulation of the phosphorylated, initiating form of Pol II (Ser5P), and SpT5 with which it associates, strongly correlates with AID binding and hypermutation (17, 67). Using ex vivo–activated B1-8 B cells, they found that this complex accumulates at V genes and switch regions in GC B cells, but only at switch regions and not at V genes in vitro. Consistent with this, the mutation frequency was even lower in the B1-8 VDJ exon in our system, which we further showed has increased APE1 expression levels relative to B6 mice. Finally, we also found that several other genes involved in SHM are expressed at higher levels in the GC than in ex vivo B cells activated in our system. Despite the low mutation frequency, our data clearly show that AID does have activity on the J_H4 V region in culture, but that most of these mutations are repaired accurately.

Similar to APE1, POLB mRNA expression is very low in GC B cells and slightly higher in culture. Interestingly, expression of the essential gene XRCC1, which coordinates interaction of APE1 and POLB, is highly increased in the GC (Fig. 5) and could promote survival where the essential proteins APE1 and POLB are limiting. Previous experiments that attempted to block the APE1/XRCC1/POLB axis in cultured B cells found that CSR was slightly increased in the absence of POLB, and in B cells that are haploinsufficient for XRCC1 (xrcc1^+/) (68–70). To explain error-prone repair, it was suggested the POLB pathway is “overwhelmed” by too many AID-induced lesions (70), but perhaps instead it is outcompeted by high levels of APE2 that recruit PCNA/TLPs instead of POLB. Similar to our results with APE1, it was recently shown that expression of POLB is very low in the GC, and suppression of POLB in CH12-F3 cells results in increased GLSμ A:T mutations (5).

Interestingly, similar to UNG, APE2 has been shown to prefer ssDNA (29), the required template of AID. The excision of abasic sites in ssDNA by APE2 would readily promote DSB formation and CSR, even without MMR. The ssDNA-binding Zf-GRF domain was not necessary for CSR or SHM in our assays, but it would be interesting to test its importance in MMR-deficient cells. Such a pathway would likely generate substantial genomic instability and therefore merits further investigation.

The authors have no financial conflicts of interest.

We thank the UMass Medical School Flow Cytometry Core Facility for expert cell sorting, and Drs. T. Honjo, T. Lindahl, D. Barnes, T. Mak, and E. Friedberg for mice. We thank Drs. Frederick Alt and Chong Wang for providing B1-8 mice and preliminary data on B1-8 SHM in culture, and Drs. V. Barretto, M. Nussenzweig, and J. Chaudhuri for retroviral constructs. We thank Drs. J. Stavnezer, R. Woodland, and J. Guikema for helpful discussions.