Efficient deamination of 5-methylcytosines in DNA by human APOBEC3A, but not by AID or APOBEC3G.

The AID/APOBEC family of enzymes in higher vertebrates converts cytosines in DNA or RNA to uracil. They play a role in antibody maturation and innate immunity against viruses, and have also been implicated in the demethylation of DNA during early embryogenesis. This is based in part on reported ability of activation-induced deaminase (AID) to deaminate 5-methylcytosines (5mC) to thymine. We have reexamined this possibility for AID and two members of human APOBEC3 family using a novel genetic system in Escherichia coli. Our results show that while all three genes show strong ability to convert C to U, only APOBEC3A is an efficient deaminator of 5mC. To confirm this, APOBEC3A was purified partially and used in an in vitro deamination assay. We found that APOBEC3A can deaminate 5mC efficiently and this activity is comparable to its C to U deamination activity. When the DNA-binding segment of AID was replaced with the corresponding segment from APOBEC3A, the resulting hybrid had much higher ability to convert 5mC to T in the genetic assay. These and other results suggest that the human AID deaminates 5mC's only weakly because the 5-methyl group fits poorly in its DNA-binding pocket.

INTRODUCTION

Deamination of cytosines in DNA to uracil has emerged as a major mechanism by which higher vertebrates protect themselves against infections. In vertebrates the enzymes that can perform this reaction is the AID/APOBEC family and one member of this family, activation-induced deaminase (AID), has an essential role in the maturation of antibodies. It diversifies the antibody repertoire by causing heavy mutagenesis of the variable segment of the rearranged antibody gene (called somatic hypermutation, SHM) or by promoting gene conversion between the variable segment and a pseudo-V segment. Additionally, AID is required for class-switch recombination (CSR) which replaces the µ constant segment of the immunoglobulin gene with other constant segments [Reviewed in (1–3)]. AID is also required for the translocation of c-myc gene to the immunoglobulin locus (4) and is implicated in the development of many cancers (5).

Among the APOBECs (APOBEC1 through APOBEC4), only APOBEC3 appears to have a protective immunity function. The APOBEC3s from a number of animals have been shown to protect cells against a number of viruses and to inhibit retrotransposition of chromosomal retroelements. APOBEC3s accomplish this through multiple mechanisms that include high level mutagenesis, strand breakage, inhibition of reverse transcription and packaging of the viral genomes. The human genome codes for seven paralogs of APOBEC3 (APOBEC3A through APOBEC3H) which contain either one or two zinc-binding motifs and in all cases the motif near the carboxy-terminus of the protein has cytosine deamination activity. In contrast non-primates contain a single APOBEC3 gene (6–8).

Morgan et al. (9) reported that purified human AID and rat APOBEC1 had the ability to deaminate 5-methylcytosines (5mC) in DNA oligomers and expression of AID in Escherichia coli also expressing the SssI methyltransferase (MTase) increased C to T mutations at a methylated cytosine in the rpoB gene. Additionally, they reported detection of expression of AID (and to a lesser extent APOBEC1) in oocytes, embryonic stem (ES) cells and other pluripotent tissues. Based on these results Morgan et al. (9) proposed that AID plays a role in epigenetic reprogramming in non-lymphoid tissues such as fertilized eggs and ES cells by causing demethylation of DNA. In its simplest form this would occur through deamination of 5mC to T by AID followed by repair of the resulting T•G mispair by base-excision repair (BER) to C:G (10,11).

In subsequent studies AID/APOBEC genes were transfected into cells and the effect of their expression on DNA demethylation was studied. In Zebra fish embryos, introduction of AID, APOBEC2a or APOBEC2b resulted in DNA demethylation, and the DNA glycosylase MBD4 enhanced this effect (12). Bhutani et al. (13) showed that during the reprogramming of human cells to induced pluripotency, siRNA-mediated inhibition of AID expression resulted in the remethylation of OCT4 and NANOG gene promoters and loss of expression of the genes. Popp et al. (14) found that primordial germ cells have significantly higher methylation at many genomic loci in AID−/− mice than in wild-type mice. Another suggested pathway for DNA demethylation involves conversion of 5mC to 5-hydroxymethylcytosine (5hmC) by Tet (Ten–eleven translocation) proteins (15), deamination of 5hmC by AID and the repair of subsequent 5-hydroxymethyluracil (5hmU)-guanine mispair by MBD4 or thymine–DNA glycosylase [TDG; (16)]. Guo et al. (17) showed that transfection of human embryonic kidney cells expressing Tet1 with AID gene reduced the level of genomic 5hmC. They also found that transfection of murine APOBEC1, human APOBEC2, APOBEC3A, APOBEC3C or APOBEC3E, but not APOBEC3B or APOBEC3G also resulted in significant reduction in genomic 5hmC (17). These results have led to the hypothesis that one or more of the AID/APOBEC family of proteins may participate in the demethylation pathways via deamination of 5mC and 5hmC (10,11,18).

Although many of these experiments are based on the assumption that AID will efficiently deaminate 5mC or 5hmC, there is significant evidence that shows that AID is not be a good deaminator of 5mC. When AID tagged with glutathione-S-transferase (GST) was expressed and purified from insect cells and used in vitro in deamination reactions, 5mC was a substantially poorer substrate than C (19,20). In one study, the rate of deamination of C in DNA oligomers was found to be ten times the rate of deamination of 5mC (20). Additionally, Kohli et al. (21) reported that when oligomers with both C’s and 5mC’s were treated with AID tagged with maltose-binding protein (MBP), C’s were deaminated much more efficiently than 5mC’s. However, the principal difficulty in making biochemically valid statements about AID is that the protein is ‘sticky’ and aggregates quickly after purification. Consequently, no studies have reported kcat/Km for AID with any substrate and quantitative assessment of substrate preferences of AID is quite difficult.

We decided to take a fresh approach to this problem and developed a simple genetic system in E. coli that can quantify deamination of 5mC or C in the same sequence context in genomic DNA. When human AID and two APOBEC3 genes were tested using this system, we found that while APOBEC3A was a strong deaminator of both C and 5mC, AID and APOBEC3G were much weaker in their ability to deaminate this modified base.

MATERIALS AND METHODS

Bacterial strains and plasmids

The kan alleles were introduced in the E. coli K-12 strain BH143 genome [Δ(mrr-hsdRMS-mcrBC) mcrA Φ80dlacZΔM15 ΔlacX74 deoR endA1 araD139 Δ(ara, leu)7697 galU galK rpsL nupG Δ (dcm-vsr)] through recombineering using the Red/ET recombination system from Gene Bridges (Heidelberg, Germany). The kan alleles in the plasmids pUP31, pUP41 and pUP44 (22) were amplified using the 70-nt primer pairs (Supplementary Table S1) each of which contained 50 nt identical to the manX gene in the chromosome. The amplification products contained the wild-type bleomycin-resistance gene in addition to the kan alleles and the recombinants were selected using zeocin (20 µg/ml) in plates. The recombinants were confirmed by DNA sequencing and the three new strains were named BH400 (from pUP31), BH300 (pUP41) and BH500 (pUP44). Escherichia coli B strain BL21DE3 was used for protein expression and purification.

The pBR322-based plasmids carrying genes for M.HpaII [pM.HpaII, (23)], Dcm [pDCM21, (24)] and M.MspI [pQ8, (25)] have been described before. The plasmid pDCM22 contains dcm+ and vsr+ genes and has also been described (24). Human AID (26), APOBEC3A(A3A) and APOBEC3G (A3G) genes were cloned into pACYC184-based plasmid pSU24 creating respectively pSUAID, pSUA3A and pSUA3G. The clones for A3A cDNA were kindly provided by Reuben Harris (University of Minnesota) and A3G cDNA was obtained from ATCC (Manassas, VA). The primers used for the cloning are listed in Supplementary Table S1. The gene for UGI was amplified from a plasmid kindly provided by Umesh Varshney (Indian Institute of Science, Bangalore, India) and inserted at an EcoRI site in pSUAID to create pSUAID-UGI. Catalytic mutants of AID (AID-E58A) and APOBEC3A (A3A-E72A) were constructed using a whole plasmid PCR mutagenesis strategy [(27), Supplementary Table S1)]. For the purification of A3A protein, both the A3A gene and the catalytic mutant A3A-E72A were amplified (Supplementary Table S1) and cloned into pET28a (+) as EcoRI-XhoI fragments. The hybrid gene AID-A3AR2 was synthesized by DNA 2.0 (Menlo Park, CA) and cloned into the pSU24 vector.

Kanamycin-resistance reversion assay

The reversion assay has been described previously (23,28). To quantify 5mC to T deaminations, AID, AID-E58A, AID-A3AR2, A3A, A3A-E72A or A3G and one of the MTase genes were coexpressed from compatible plasmids and kanamycin-resistant revertants (50 µg/ml, phenotype- KanR) were scored. KanR revertant frequency is the ratio (Number of KanR revertants/Total number of viable cells). To quantify C to U deaminations, either the UGI gene was expressed from the same plasmid as AID or one of the APOBECs, or an ung strain was used as host. To study repair of T•G mispairs, the plasmid pDCM21 (dcm+) or pDCM22 (dcm+ vsr+) was introduced in BH400 and the KanR revertant frequencies were determined.

To determine the base change in the kan gene, 12 independent revertants were amplified, the products purified using PCR purification kit (Epoch-Life Science) and sequenced (Supplementary Table S1). The sequences of revertants were aligned with the sequences of the kan alleles using MacVector software (MacVector, Inc., Cary, NC) and the mutations were identified.

Uracil quantification assay

The uracils in the genomic DNA were quantified as described previously (29). Briefly, the genomic DNA was incubated with methoxyamine to block any pre-existing abasic sites and was treated with E. coli UDG (New England Biolabs) and aldehyde-reactive probe (Dojindo Molecular Technologies, Rockville, MD). The DNA was transferred to a nylon membrane (EMD-Millipore, Billerica, MA) and cross linked to the membrane using a UV light. The membrane was incubated with 5 × 10−4 mg/ml of streptavidin-Cy5 (GE Healthcare), it was scanned using a Typhoon 9210 phosphorimager (GE Healthcare) and the images were analyzed with ImageQuant software. The uracil standard was a 75-mer duplex containing a single U•G mispair.

Purification of A3A and its mutant

BL21DE3 cells expressing A3A or A3A-E72A gene in pET28a (+) were grown to mid-log phase and transcription was induced with the addition of IPTG. The cells were harvested after 5 h and broken using the French Pressure Cell Press (Thermo Spectronic) and the cell-free lysate was cleared by centrifugation. The lysate was passed over a Ni–NTA column (Novagen, Madison, WI) and the bound proteins were eluted with 250 mM imidazole. The eluted proteins were dialyzed and concentrated using Amicon Ultra Centrifugal devices (EMD-Milipore, Billerica, MA). The concentrated proteins were equilibrated in the storage buffer (20 mM Tris–HCl, 50 mM NaCl, 1 mM EDTA, 1 mM DTT, 10% glycerol).

Biochemical assay for C and 5mC deamination

The substrates for the study of C and 5mC deamination, A3A–C and A3A–5mC respectively, are listed in Supplementary Table S2. Six picomoles of oligomer A3A–C was incubated at 37°C with 140 ng of partially purified A3A enzyme in a 10 µl volume in the reaction buffer [40 mM Tri–HCl (pH 7.5), 5 mM EDTA, 1 mM DTT, 40 mM NaCl]. The reaction was terminated by the addition of 1,10-phenanthroline (Sigma-Aldrich) to 5 mM. Two units of E. coli UDG (New England Biolabs, Ipswich, MA) were added to the reaction and incubation was continued at 37°C for 1 h. The reactions were stopped by adding NaOH to 0.1 M and heating to 95°C for 7 min. The products were separated on a 15% denaturing acrylamide gel and the gel was scanned using Typhoon 9210 phosphorimager. ImageQuant software was used to quantify the intensities of the substrates and the deaminated products. For 5mC deamination 2 pmol of oligomer A3A–5mC was incubated with 140 ng of partially purified protein at 37°C for 1 h. The oligomer complement added to the reaction at 3-fold molar excess to create a T•G mismatch. The duplex was incubated with 1.5 units of thermostable thymine DNA glycosylase (Trevigen, Gaithersburg, MD) for 1 h at 47°C and the products were processed and analyzed in a manner similar to C-deamination products.

RESULTS

Construction and validation of genetic system for 5mC deamination

We described previously two defective alleles of the kanamycin-resistance gene (kan) that can be methylated in E. coli by different MTases and used to quantify 5mC to thymine conversions (23). In this plasmid-based system, the kan− alleles revert to kan+ (phenotype-KanR) through 5mC to T deamination increasing the KanR frequency by at least an order of magnitude (23). We have expanded and adapted this system for the study of 5mC to T conversion by AID/Apobec family proteins.

To reduce the number of plasmids maintained in the cells, we constructed three strains of E. coli, each with a different kan allele (22,23) inserted into the manX gene using homologous recombination (Supplementary Figure S1). Three different cytosine MTase genes were introduced into these E. coli strains (BH300, BH400 and BH500) on plasmids to create five different sequence contexts for cytosine methylation. These included two in which the 5mC was in a CpG sequence context while in one it was in the WRC context (W is A or T, R is purine) preferred by AID (30). The sequence context of 5mCs in the different strains is shown in Figure 1A.

Figure 1.

(A) Sequence context of 5mC in kan alleles. The five different methylation contexts created in the kan gene through changes in base sequence or methylation are shown. Three different kan alleles are present in E. coli strains BH300, BH400 and BH500. The gene for the MTase M.HpaII, M.MspI or Dcm was introduced in these strains to methylate one of the cytosines in a proline codon (underlined in the figure). The methylated cytosine is indicated by ‘me’ above the C. The five bases unique to each sequence context are indicated by a bracket below the sequence. 5mC to thymine mutation changes the proline codon to either leucine or serine and changes the cellular phenotype from kanamycin-sensitive (KanS) to kanamycin-resistant (KanR). (B) Effects of MTases on KanR revertant frequency. The frequency of revertants with or without the presence of a MTase is shown. The bacterial host used (BH300, BH400 or BH500) and the relevant methylated sequences are shown. The genes for M.HpaII, M.MspI or Dcm were introduced in the cells to methylate the DNA. ‘M’ is 5-mC and the horizontal line within the data points is the mean value.

We first confirmed that the chromosomal DNA in these strains was appropriately methylated. The genomic DNA from cells expressing Dcm, M.HpaII or M.MspI MTases was digested respectively with EcoRII, HpaII or MspI and was found to be resistant to the endonucleases (Supplementary Figure S2). We also compared the frequency of spontaneous KanR revertants in the presence and absence of MTases. In each case, the frequency of KanR revertants increased by a factor of ∼100 (Figure 1B) and this result is consistent with the previously reported increases in KanR revertant frequencies due to methylation by Dcm and M.HpaII (23,28). The large magnitude of the increase in revertant frequency is due partly to the fact that 5mC deaminates at approximately four times the rate of deamination of C (31) and partly because of the inability of these cells to repair T•G mispairs created by 5mC deamination. E. coli lacks DNA glycosylases like MBD4 and TDG that excise thymines from a T•G mispair and the gene for the T•G-specific endonuclease, Vsr, has been deleted from the genomes of the cells used (see ‘Materials and Methods’ section, and below). We also sequenced 12 independent KanR revertants from cells that expressed different MTases and nearly all the revertants had the methylated cytosine changed to thymine (Supplementary Table S3). The near complete absence of mutations at unmethylated cytosines is probably because the E. coli strains BH300, BH400 and BH500 are proficient in the repair of U•G mispairs created by the deamination of C. These results show that in this genetic system the frequency of KanR revertants is an accurate measure of the 5mC to T conversions.

Finally, we also confirmed that a majority of the revertants arose through the conversion of a 5mC:G pair to a T•G mispair in the case of at least the Dcm MTase. We did this by introducing the gene vsr+ in the strain BH400 in addition to Dcm. Vsr is a T•G-specific endonuclease that hydrolyzes the phosphodiester linkage immediately upstream of the mispaired T (32) initiating very short-patch (VSP) repair pathway that replaces the T with a C [Figure 2A; (33)]. As expected, when dcm+ gene alone was introduced in E. coli, KanR frequency increased by a factor of ∼55 compared to the vector control, but when both dcm+ and vsr+ genes were introduced in the cells, the increase was reduced to only ∼4-fold over the control (Figure 2B). Together these results show that the genetic reversion system used here can be used to quantify 5mC to T deaminations.

Figure 2.

(A) Very short-patch repair of T•G mispairs in E. coli. The pathway (33) by which E. coli VSP repair replaces T•G mispairs created by the deamination of 5mCs in the Dcm sequence context with C•G is shown. Deamination of one of the 5mCs in a Dcm sequence context creates a T•G mispair. The Vsr endonuclease hydrolyzes the phosphodiester linkage 5′ of the mispaired T and DNA polymerase I (PolI) and DNA ligase complete the reaction. (B) KanR revertant frequency with and without VSP repair. The revertant frequencies in the presence of Dcm alone or with both Dcm and Vsr in the cells (BH400) are shown and are compared with the frequency in the absence of either enzyme. The horizontal line within the data points is the median value.

Modest deamination of 5mC by human AID

When human AID gene was introduced into BH500 lacking DNA methylation, ∼1 in 106 cells became resistant to kanamycin (Figure 3A). However, BH500 is proficient in the repair of U•G and hence bulk of the uracils generated by cytosine deamination are expected to be repaired. To quantify the full extent of cytosine deaminations caused by AID, we coexpressed Ung inhibitor UGI (34) in the cells. Expression of both AID and UGI increased the frequency of KanR revertants ∼200-fold to >10−4 (Figure 3A). This result is consistent with previous reports that human AID is a strong deaminator of cytosines in the E. coli genome (35).

Figure 3.

(A) Cytosine deamination by AID. KanR revertant frequencies in BH500 cells expressing AID alone or AID and UGI. The horizontal line within the data points is the median value. (B) 5mC deamination by AID. KanR revertant frequencies in BH500 cells expressing AID, E58A mutant of AID or empty vector are shown. The horizontal line within the data points is the median value. (C) Quantification of genomic uracils. The amount of genomic uracil created by AID alone or AID and UGI is shown. The error bars indicate the standard deviation.

In contrast, when M.MspI was present in BH500 along with AID, only a modest increase in 5mC deamination was scored by the KanR assay. The median frequency of revertants was 10−5, which was only 1.9-fold higher than the frequency due to empty vector and only slightly higher (1.5-fold) than a catalytically inactive mutant of AID (Figure 3B). It should be noted that the 5mC in this genetic assay was in the WRC sequence context preferred by AID (30) and the overall KanR frequency was highest in this sequence context compared to other contexts (Supplementary Figure S3). Consistent with the results in the WRC context, only modest increases in the KanR frequency were seen due to AID in these other contexts including two where the methylated base was in a CpG dinucleotide. In no case was the increase in revertant frequency >2-fold compared to the vector control (Supplementary Figure S3A–D).

We wanted to make sure that in cells where AID was a poor deaminator of 5mC, it was still a strong deaminator of cytosines. To establish this, we used a biochemical assay for the quantification of genomic uracils. The assay introduces a fluorescent Cy5 tag at sites in DNA that have uracils, and uses Cy5 fluorescence to quantify the uracils and has been described previously (36–38). The results show that while presence of AID caused 5mC to T conversion to increase <2-fold (Figure 3B), the amount of uracils in the genome of same cells increased ∼10-fold (Figure 3C). These results show that while AID acts as a strong deaminator of cytosines, it is poor at deaminating 5mC.

Efficient 5mC deaminations by APOBEC3A, but not APOBEC3G

APOBEC3A (A3A) and APOBEC3G (A3G) are two members of the human APOBEC3 family and are sequence homologs of AID. First, we expressed the full-length A3A and A3G in ung− E. coli and compared their ability to deaminate cytosine to that of AID. As expected, A3G was significantly better at deaminating cytosines in its preferred sequence context (a run of C’s) than AID (Figure 4A). However, when A3G was tested for its ability to deaminate 5mC in the preferred sequence, no increase in KanR revertant frequency was detected (Figure 4B). In this sequence context, AID is also inefficient at deaminating 5mC (Figure 4B and Supplementary Figure S3A). In these and other experiments we have been unable to detect any deamination of 5mC by A3G (data not shown).

Figure 4.

(A) Comparison of cytosine deamination by AID and A3G. KanR revertant frequencies in BH300 cells expressing AID, E58A mutant of AID or A3G are shown. The horizontal line within the data points is the median value. (B) Comparison of 5mC deamination by AID and A3G. KanR revertant frequencies in BH300 cells expressing vector, AID or A3G along with M.HpaII are shown. The horizontal line within the data points is the median value.

In contrast, A3A was not only a much stronger mutator than AID when the genetic system was designed to score C to U deaminations (Figure 5A), but also when 5mC to T conversions were scored (Figure 5B). The methylated cytosine in this experiment was in the CpG context and a cytosine flanked the CpG on the 5′-side. We chose this sequence context for experiments with A3A because this enzyme is known to prefer a T or a C on the 5′-side of the target cytosine (39). We also found A3A to increase substantially the revertant frequency when 5mC was in CpH context (H is not G; data not shown). Changing a conserved glutamate residue in A3A expected to be required for catalysis to alanine completely eliminated 5mC to T mutations (Figure 5C). When independent revertants obtained in experiments where cells expressed an MTase along with AID or A3A were sequenced, an overwhelming majority of the mutations were at the methylated cytosines (Supplementary Table S4). This shows that the mutations resulted from a 5mC to T change. The presence of WT A3A in cells increased the revertant frequency by at least 15-fold compared to empty vector and in some experiments it was much higher (Figure 5C and data not shown). These data suggest that A3A is much more efficient at deaminating 5mC than AID.

Figure 5.

(A) Comparison of cytosine deamination by AID and A3A. KanR revertant frequencies in BH500 cells expressing AID, E58A mutant of AID or A3A are shown. The horizontal line within the data points is the median value. (B) Comparison of 5mC deamination by AID and A3A. KanR revertant frequencies in BH500 cells expressing AID, E58A mutant of AID or A3A along with M.HpaII are shown. (C) 5mC deamination by A3A. KanR revertant frequencies in BH500 cells expressing A3A, E72A mutant of A3A or empty vector are shown. The horizontal line within the data points is the median value.

5mC to thymine deaminations by APOBEC3A in vitro

To confirm this deamination activity biochemically, the human A3A gene was modified at its 3′-end with six His codons and the tagged protein was purified partially over a nickel affinity column. The purified protein was active and was able to completely convert a cytosine in a synthetic oligomer to uracil (Figure 6A, lane 4). Based on the purity of the enzyme (Supplementary Figure S4), we conclude that ∼2 pmol of the enzyme completely converted 6 pmol of C’s to U’s in about one hour. These data suggest that the enzyme acts slowly on the substrate used, but it does turn over.

Figure 6.

(A) Cytosine and 5mC deamination by A3A. Fluorescently labeled DNA oligomers with a single C or 5mC were treated with A3A or A3A-E72A. The C-containing oligomer was treated with UDG and cleaved while the 5mC-containing oligomer was hybridized to its complement, treated with TDG and cleaved. (B) Kinetics of cytosine and 5mC deamination by A3A. Fluorescently labeled DNA oligomers with a single C or 5mC were treated with A3A and the reactions were stopped at indicated times. The complementary strand was hybridized to the oligomer and UDG or TDG was added to cleave the DNA strand. DNA duplex containing U•G or T•G mispairs respectively serve as controls for the efficiency of reactions with UDG and TDG. (C) Quantification of data in part B above. The data are normalized for the efficiency of UDG and TDG reactions, and the signal at zero time was subtracted from all data points.

The partially purified A3A also converted 5mC to T efficiently. To demonstrate this, an oligomer with a single 5mC was treated with this enzyme and an excess of the complementary strand was added to create T•G mispairs at the deaminated 5mCs. This duplex was successively treated with the TDG and NaOH to create strand breaks. Using this procedure we found that under similar reaction conditions A3A treatment converted ∼78% of 5mC containing substrate to product (Figure 6A, lane 8), while converting ∼99% of C containing substrate to product (Figure 6A, lane 4). As expected, A3A mutant with E72A mutation was unable to convert 5mC to T (Figure 6A, lane 10). When the time-dependence of deamination by A3A was studied, significant conversion of 5mC could be detected even at the earliest time points (Figure 6B and C). For technical reasons, there was some set-to-set variation in the data (see ‘Discussion’ section), but in some experiments >90% of 5mC was converted to T in 60–120 min (Supplementary Figure S5, Set 2 and data not shown). Thus, both genetic and biochemical data show that A3A is an efficient deaminase of both C and 5mC in DNA.

An eight amino acid segment of A3A confers AID 5mC deamination ability

We and others showed previously that when the putative DNA-binding domain (DBD) of AID is replaced with segment from either A3G or A3F carboxy-terminal domain, the sequence preference of AID is altered to that of the latter enzymes (21,40,41). We hypothesized that A3A may contain a DBD that can accommodate 5mC and replacement of the DBD of AID with this domain may allow AID to deaminate 5mCs. To test this, the DBD of AID was replaced with the corresponding sequences from A3A and the resulting hybrid (AID-A3AR2; Figure 7A) was tested for 5mC deamination ability in the genetic assay. The hybrid gene caused much higher KanR revertant frequency than AID and was almost as efficient as A3A at deaminating 5mC in this assay (Figure 7B). Thus, an eight amino acid segment of A3A was able to confer upon AID the ability to deaminate 5mC in DNA.

Figure 7.

(A) Domain swap between AID and A3A. The sequence of the putative DBDs of AID (21,40,41) and A3A are shown schematically. The DBD of A3A was identified by aligning the sequence of this protein with the sequence of AID and of the carboxy-terminal domain of A3G. AID-A3AR2 contains all of AID except its DBD which is replaced with eight amino acid DBD from A3A. The numbers above and below the sequences are amino acid residue numbers. (B) Comparison of 5mC deamination by AID, A3A and AID-A3AR2. KanR revertant frequencies in BH500 cells expressing the different proteins are shown. The horizontal line within the data points is the median value.

DISCUSSION

We showed here that while human AID and A3G proteins were quite proficient at converting C to U in E. coli genomic DNA, they had little or no activity when 5mC was the substrate. Furthermore, the strong cytosine deamination activity by AID (Figure 3A and C) in the same cells in which little 5mC to T conversion could be detected (Figure 3B and Supplementary Figure S3A–D), strongly argues that the poor 5mC deamination activity was not due to low expression level of the protein or its stability in E. coli. The genetic system used here scored 5mC to T mutations in the WRC sequence context preferred by AID, as well as in the CpG context where the bulk of 5mC is found in mammalian cells. In no case was the increase in 5mC to T mutations >2-fold above the vector control (Figure 3B and Supplementary Figure S3A–D). These results are inconsistent with the conclusions of the largely qualitative studies reported by Morgan et al. (9) and models of DNA demethylation that depend upon deamination of 5mC by AID (10,11).

We were concerned that the low 5mC deamination abilities of AID and A3G scored by the KanR reversion assay reflected some inherent weakness of the genetic system. To dispel such criticisms, we tested two additional APOBEC3 family members, A3A and APOBEC3C (A3C) with this assay. The expression of A3C was toxic to E. coli and this gene did not give consistent results in the KanR assay (data not shown). However, A3A was more mutagenic than AID in all sequence contexts including a WRC sequence in which a C to U deamination was being scored (Figure 5A). Previous studies using the E. coli rifampicin-resistance assay showed that A3G is somewhat more mutagenic than AID (42) and other studies using the same assay have shown A3A to be more mutagenic than A3G (43,44). While these earlier results when taken together suggested that A3A was a more potent mutator than AID, the present study is the first one to compare directly these two deaminases. Importantly, A3A also scored much better at deaminating 5mC than AID (Figure 5C). When M.HpaII is expressed in E. coli, the genome contains ∼64 cytosines for every 5mC and hence the high frequency deamination of one 5mC in the proline codon in the kan gene in a sea of C’s suggests that A3A does not have a strong preference for C over 5mC.

We confirmed this biochemically by purifying partially A3A from E. coli and testing it in vitro for C and 5mC deamination using single-stranded DNA substrates. A3A was able to deaminate both the bases with only a moderate preference for C over 5mC. It should be noted that the efficiency of the TDG reaction varied from experiment to experiment partly because the DNA duplex was unstable at the recommended reaction temperature (47°C). Consequently, not all the T•G mispairs created by A3A are processed by TDG and hence there is a lower than actual appearance of 5mC to T conversion in Figure 6B and Supplementary Figure S5. Regardless, these data confirm the conclusion from our genetic studies that A3A is an efficient deaminase of 5mC. The only quantitative biochemical study of AID using DNAs with C and 5mC concluded that the former base is ∼10-fold better substrate than latter (20). Our data show (Figure 6 and Supplementary Figure S5) that cytosine is preferred over 5mC by A3A by a factor of only two to three. It should be noted that there has been no previous report of any enzyme from any organism that efficiently convert 5mC in DNA to T.

It can be argued that the reason why AID behaves as a weak deaminator of 5mC in our E. coli assays is because the protein lacks a key modification or partner. While it is not possible to disprove this, it is unlikely for several reasons. First, studies have shown that AID is a strong mutator at unmethylated cytosines in E. coli and our genetic system reproduces the dependence of this activity on transcription of the target gene in the same way as in B lymphocytes (3). Second, human AID purified from insect cells contains the only modification of the protein [Ser-38 phosphorylation; (45)] that is thought to affect its substrate specificity (46). This phosphorylated protein was 10-fold less active against 5mCs’ than Cs’ (19,20). Third, we have shown here that AID can be changed into a more active 5mC deaminase by replacing its putative DBD with the corresponding domain of A3A (Figure 7B). This suggests that while A3A is able to accommodate the 5-methyl group on the target cytosine in its active site, AID may not be able to do so.

We attempted to dock 5-methyl-dC nucleotide into the active site of A3G using Autodock Vina (http://autodock.scripps.edu/) and found that Tyr-315 clashed with the methyl group (M. Carpenter and A.S. Bhagwat, unpublished results). Homology modeling of AID and A3A based on the published structures of A3G suggest that while this conserved tyrosine is similarly positioned in AID, it is much further away from the cytosine-binding pocket in A3A [(47) and data not shown]. Thus, the position of this tyrosine in the active sites of AID and A3G may be the principal reason for their poor ability to deaminate 5mC.

Our demonstration that A3A is a strong 5mC deaminase does not necessarily mean that this enzyme plays a significant role in DNA demethylation during early embryogenesis. While A3A is imported into the nucleus (6), its expression has not been reported in germ cells or embryonic cells. It has been shown to play a role in restricting human viruses and retrotransposons (39,48), and in degrading foreign DNA (49). There is some evidence that it may act on 5mCs in DNA, but the evidence is indirect and comes from somatic cells. In experiments in which A3A gene was transfected into human cell lines, several C to T or G to A mutations were detected in CpG sequences within c-myc and p53 genes suggesting 5mC to T deaminations (50). However, this study did not determine the state of methylation of CpGs in these genes and hence it is possible that these were merely C to T changes.

There are several biochemical arguments against AID, A3A or other cytosine deaminases playing a major role in the demethylation of DNA in pluripotent cells. First, cytosine is the preferred substrate for all AID/APOBEC family deaminases for which this has been studied. As there is ∼30-fold excess of unmethylated cytosines in mammalian genomic DNA over 5mC, it is difficult to visualize how most 5mCs could be deaminated without also deaminating all cytosines. Second, all AID/APOBEC family deaminases prefer single-stranded DNA as substrate. It is difficult to see how all the genomic DNA can be presented to these enzymes in this form in the paternal genome where demethylation is thought to occur prior to first round of replication (51). Third, a replacement of all 5mCs in the genome with cytosines through BER would require several million separate repair events per cell. The bulk of BER uses DNA polymerase β which has an error frequency of ∼1 in 104 bases synthesized (52). This would lead to an unacceptably high mutational load on the embryonic genome. In summary, the work presented here suggests that the ability of human A3A to deaminate 5mC may be biological relevant, however much more work is needed before a specific biological role can be ascribed to this enzymatic activity.

SUPPLEMENTARY DATA

Supplementary Data are available at NAR Online: Supplementary Tables 1–4 and Supplementary Figures 1–5.

FUNDING

National Institutes of Health (NIH) [GM 57200 and CA 153936]. Funding for open access charge: NIH GM57200.

Conflict of interest statement. None declared.