Evidence for Editing of Human Papillomavirus DNA by APOBEC3 in Benign and Precancerous Lesions

See allHide authors and affiliations

Science  11 Apr 2008:
Vol. 320, Issue 5873, pp. 230-233
DOI: 10.1126/science.1153201


Cytidine deaminases of the APOBEC3 family all have specificity for single-stranded DNA, which may become exposed during replication or transcription of double-stranded DNA. Three human APOBEC3A (hA3A), hA3B, and hA3H genes are expressed in keratinocytes and skin, leading us to determine whether genetic editing of human papillomavirus (HPV) DNA occurred. In a study of HPV1a plantar warts and HPV16 precancerous cervical biopsies, hyperedited HPV1a and HPV16 genomes were found. Strictly analogous results were obtained from transfection experiments with HPV plasmid DNA and the three nuclear localized enzymes: hA3A, hA3C, and hA3H. Thus, stochastic or transient overexpression of APOBEC3 genes may expose the genome to a broad spectrum of mutations that could influence the development of tumors.

Human APOBEC3 molecules deaminate cytidine residues in single-stranded DNA (ssDNA) and have been demonstrated to have antiviral effects (15). Human immuno-deficiency virus–1 (HIV-1) cDNA in particular is vulnerable to the action of the cytoplasmic APOBEC3F and APOBEC3G cytidine deaminases (hA3F and hA3G) (6, 7). Of the seven-gene cluster on chromosome 22, hA3A, hA3C, and hA3H are mainly nuclear, whereas hA3B is both nuclear and cytoplasmic (8, 9). Human A3A and hA3B are expressed in psoriatic keratinocytes, and hA3A is up-regulated in acne lesions and can be induced by phorbol 12-myristate 13-acetate (10, 11). Incidentally, hA3H is also expressed in normal skin (12, 13). We hypothesized that the DNA of human papillomaviruses, which replicate in cutaneous and mucosal keratinocytes, might be vulnerable to editing by some of the nuclear A3 deaminases.

In light of the predominant APOBEC3 expression data in cutaneous keratinocytes, total DNAwas extracted from six HPV1a-positive plantar warts. For mutational analysis, a region corresponding to the origin of replication/promoter region was selected, because it seemed likely that this region might exist in a single-stranded state more frequently than any other region of the HPV genome. Differential DNA denaturation polymerase chain reaction (3D-PCR) was used to selectively amplify AT-rich edited genomes (14, 15). This technique relies on the fact that AT-rich DNA denatures at a lower temperature than GC-rich DNA and has proven successful at selectively amplifying APOBEC3-edited retroviral cDNA (16). For one out of the six plantar warts, monotonously G→A or C→T substituted HPV1a sequences were identified (Fig. 1, A and B), meaning that both strands had undergone editing.

Fig. 1.

APOBEC3 editing of HPV1a genomes in vivo and in vitro. (A) A selection of G→A and C→T edited HPV1a sequences derived from a plantar wart (sample HPV1a-1). Only sequence differences are noted with respect to the viral plus strand. For clarity, only the most 5′ 120 bases of the 315–base pair (bp) segment are shown. The numbers to the right indicate the total number of edited bases per sequence. (B) Mutation matrices for the edited HPV1a genomes in vivo and in vitro. The number of bases sequenced is shown below each matrix. (C) 3′ Dinucleotide context analysis of in vivo edited HPV sequences. The cytidine target is underlined. The expected values were calculated for both strands and weighted by the number of G→A and C→T edited genomes. (D) 5′ Dinucleotide context analysis of G→A hypermutated HPV genomes in vivo and in vitro.

Averaging across both strands, we determined that the mean cytidine editing frequency was ∼11% (range: 4 to 61% per clone). Within the nucleus, G:U mismatch repair, which predominantly yields the original G:C, will mitigate against the effects of APOBEC3 editing. Hence, it is possible that a vestigial C→T transition reflects the opposing forces of APOBEC3 editing and G:U mismatch repair. Analysis of the dinucleotide context of edited sites showed that there was no pronounced 3′ nucleotide context (Fig. 1C), such as CpG, thus ruling out a cytidine hypermethylation/deamination-related phenomenon. By contrast, there was a strong 5′ effect favoring TpC and CpC, typical of some APOBEC3 deaminases (17) (Fig. 1D).

Based on this evidence for in vivo editing of HPV sequences, we next tested the potential for APOBEC3 editing of HPV DNA in vitro. Human embryonic kidney–293T cells were cotransfected by HPV1a plasmid DNA along with each of six APOBEC3 genes: hA3A, hA3B, hA3C, hA3F, hA3G, and hA3H. At 72 hours, total DNA was recovered and 3D-PCR was performed. When compared with the HPV1a plasmid plus vector alone transfection [denaturation temperature (Td)= 84.6°C], 3D-PCR products were recovered at a lower Td (82.0°C) from the hA3A, hA3C, and hA3H cotransfections. Sequencing of cloned 3D-PCR products showed extensive and monotonous cytidine deamination of both DNA strands (Fig. 1, A and B). The mean cytidine editing frequencies in vitro, 25 to 29% (range: 13 to 60% per clone), were about twofold higher than those in vivo, which may reflect stronger APOBEC3 gene expression driven by the powerful cytomegalovirus immediate-early promoter. The ssDNA cytidine editing activity reported here for human A3H appears to be just as strong as that seen for hA3A and hA3C. Furthermore, the dinucleotide context of edited sites for all three hA3 deaminases showed a clear 5′ preference for TpC and CpC and a strong aversion for ApC that correlates well with the observed editing contexts in vivo (Fig. 1D). Because the preferred dinucleotide context for human activation-induced deaminase (AID) is ApC and GpC (18, 19), it seems probable that this enzyme can be excluded from the list of potential mutators.

Analysis of site-specific editing frequencies, which factors in cold spots, revealed significant correlations (P <10–4) between the in vivo data set and all three in vitro data sets. Indeed, as hA3A and hA3H are expressed in keratinocytes, whereas hA3C is probably the most widely expressed of all the human A3 deaminases (20), there is no reason to exclude one from the others. Cotransfection experiments with plasmids encoding human APOBEC1 and APOBEC2 (hA1 and hA2), or the orthologous murine genes mA1, mA2, and mA3, failed to yield hyperedited HPV1a genomes. Given the negative result for hA3B and the fact that hA1 and mA1 traffic between the cytoplasm and nucleus, a nuclear localization perse is apparently insufficient for deamination of HPV1a DNA. These findings extend the range of viral targets for APOBEC3 enzymes to include double-stranded DNA viruses.

Human papillomaviruses can be generally divided into those that infect cutaneous or mucosal keratinocytes, a feature strongly reflected in their phylogenic clustering (21). Among the latter group are found the HPVs that are strongly associated with cervical cancer. In an extensive European study, ∼65 and ∼6% of cervical cancers were associated with HPV16 and HPV18, respectively (22), with at least 11 other strains also showing a strong association (23).

Although it is unclear whether APOBEC3 genes are expressed in mucosal keratinocytes, the evidence for APOBEC3 editing of HPV1a genomes led us to examine total DNA from nine HPV16-positive precancerous cervical biopsies. 3D-PCR was again used to amplify a comparable region spanning the origin of replication/promoter region. Cloning and sequencing of the last positive points of sample HPV16-29 yielded 19 sequences with monotonously G→A or C→T transitions, which is indicative of editing of both DNA strands (Fig. 2A). A single C→Tedited sequence was also recovered from a second HPV16 sample (HPV16-33, Fig. 2B). A cutoff of three or more C→T transitions was imposed to reduce the impact of natural variation and PCR error in designating APOBEC3 editing. Nonetheless, an analysis of all the remaining HPV16 genomes recovered by 3D-PCRshoweda netexcessof G→Aand C→T transitions (Fig. 2, A to C). The mean editing frequency was ∼9% (range: 5 to 24% per clone). Just as for HPV1a in vivo, the preferred dinucleotide context for editing was TpC and CpC, suggesting that hA3A, hA3C, and hA3H were involved (Fig. 2D).

Fig. 2.

APOBEC3 editing of HPV16 genomes in vivo. (A) A selection of G→A and C→T edited HPV16 sequences derived from precancerous cervical biopsies from sample HPV16-29. Only sequence differences are noted with respect to the viral plus strand. For clarity, only the most 5′ 225 bases of the 325-bp segment are shown. The numbers to the right indicate the total number of edited bases per sequence. (B) C→T hypermutated sequence derived from precancerous cervical biopsy HPV16-33, the complete region being shown. The unedited reference sequence for samples 29 and 33 is identical to that of AF067024. (C) Mutation matrices for the edited HPV16 genomes in vivo. The number of bases sequenced is shown below the matrix. (D) 5′ Dinucleotide context analysis of hyperedited HPV16 genomes in vivo. (E) Neighbor-joining tree based on the protein sequences of five human single-domain cytidine deaminases including hA2.

Transcriptionally active HPV genomes are in the form of minichromosomes, complete with nucleosomes (24). If they are as the data suggest (i.e., vulnerable to cytidine deamination, say in a subset of cells with overexpressed hA3A, hA3C, and hA3H), then it raises the question as to how chromosomal DNA is normally protected from these three powerful deaminases (Figs. 1 and 2). AID is a cytidine deaminase that mediates class switch recombination and somatic hypermutation of rearranged immunoglobulin genes (25, 26). Not only does this gene have an exon-intron structure similar to that of hA3A, hA3C, and hA3H, but human AID also falls into a cluster with hA3A and hA3C, showing ∼43 to 46% amino acid identity (Fig. 2E). By contrast, hA3H shows only 34 to 37% amino acid identity with members of this AID/hA3A/hA3C cluster. Ec-topic expression of AID may be found in a variety of cell types including nonlymphoid tumors, again suggesting a link between cytidine deamination and tumorigenesis (27). As the predominant mutation in cancer genomes is the C→T transition (28, 29), irrespective of CpG, the present findings are consistent with the possibility that stochastic or transient overexpression of either of four human cytidine deaminase genes— AID, hA3A, hA3C, and hA3H—might be sufficient to create an initial broad mutant spectrum from which the cancer genome eventually emerges.

Supporting Online Material

Materials and Methods

References and Notes

View Abstract

Stay Connected to Science

Navigate This Article