The human APOBEC (apolipoprotein B mRNA editing catalytic polypeptide-like) family includes enzymes that catalyze the hydrolytic deamination of cytidine to uridine or deoxycytidine to deoxyuridine. This family is composed of eleven known members: APOBEC1, APOBEC2, APOBEC3 (further classified as A3A to A3H), APOBEC4, and AID (activation induced deaminase). APOBEC proteins are associated with several functions involving editing of DNA or RNA (reviewed by Smith et al ¹ ). APOBEC1 mediates deamination of cytidine at position 6666 of apolipoprotein B mRNA, resulting in the introduction of a premature stop codon and the production of the short form of the protein ^{2– 4} . APOBEC2 is essential for muscle tissue development ⁵ . APOBEC4 has no ascribed function so far ⁶ . AID deaminates genomic ssDNA of B cells, initiating immunoglobulin somatic hypermutation and class switch processes ^{7– 9} . Most notably, APOBEC3 enzymes participate in innate immunity against retroviruses and endogenous retroelements ^{10– 12} . Sheehy et al. demonstrated that A3G also plays a role in immunity against human immunodeficiency virus type 1 (HIV-1) ¹³ . For its antiviral role, A3G is packaged along with viral RNA ¹⁴ . Upon infection of target cells and during the reverse transcription process, A3G deaminates the cytosine residues of the nascent first retroviral DNA strand into uraciles. The resulting uracil residues serve as templates for the incorporation of adenine, which at the end result in strand-specific C/G to T/A transitions and loss of infectivity through lethal mutagenesis ^{15– 19} . On the other hand, sub-lethal mutagenic activity of APOBEC3 proteins may end up being an additional source for HIV-1 genetic diversity, hence bolstering its evolvability ^{20– 22} . APOBEC3 proteins have been shown to inhibit other retroviruses (simian immunodeficiency virus ²³ , equine infectious anemia virus ²⁴ , foamy virus ²⁵ , human T-cell leukemia virus ²⁶ , and murine leukemia virus ²⁷ ), pararetroviruses (hepatitis B virus ²⁸ ) and DNA viruses (herpes simplex virus 1 ^{29, 30} , Epstein-Barr virus ³⁰ , HSV-1 and EBV respectively, and human papillomavirus ³¹ ). In the cases of HSV-1 and EBV, the antiviral role of deaminases has not yet been demonstrated ³⁰ . Evidence also exists that A3G significantly interferes with negative-sense RNA viruses lacking a DNA replicative phase ³² . For example, the transcription and protein accumulation of measles virus, mumps virus and respiratory syncytial virus (RSV) was reduced 50–70%, whereas the frequency of C/G to U/A mutations was ~4-fold increased after overexpressing A3G in Vero cells ³² . In contrast, A3G plays no antiviral activity against influenza A virus despite being highly induced in infected cells as part of a general IFN-β response to infection ^{33, 34} .

Human APOBEC belongs to a superfamily of polynucleotide cytidine and deoxycytidine deaminases distributed throughout the biological world ³⁵ . All family members contain a zinc finger domain (CDD), identifiable by the signature (H/C)-x-E-x25-30P-C-x-x-C. Plants are not an exception and, for example, the Arabidopsis thaliana genome encodes nine putative cytidine deaminases (with genes named AtCDA1 to AtCDA9). Whilst the AtCDA1 gene is located in chromosome II, the other eight genes are located in chromosome IV. In the case of rice and other monocots, only one CDA has been identified ³⁵ . Interestingly, this CDA expression was highly induced as part of the general stress response of rice against infection of the fungal pathogen Magnaporthe grisea, resulting in an excess of A to G and U to C mutations in defense-related genes ³⁶ . Edited dsRNAs might be retained in the nucleus and degraded, generating miRNAs and siRNAs ³⁷ . Given the relevance of deamination as an antiviral innate response in animals, we sought first to determine whether any of the AtCDA proteins encoded by plants can participate in deaminating the genome of the pararetrovirus, cauliflower mosaic virus (CaMV; genus Caulimovirus, family Caulimoviridae) and, second, we sought to explore whether this deamination may negatively impact viral infection. We hypothesize that deamination may take place mainly at the reverse transcription step. The CaMV genome is constituted by a single molecule of circular double-stranded DNA of 8 kbp ³⁸ . The DNA of CaMV has three discontinuities, Δ1 in the negative-sense strand (or a strand), and Δ2 and Δ3 in the positive-sense strand (yielding the b and g strands). In short, the replication cycle of CaMV is as follows ³⁸ : in the nucleus of the infected cell, the a strand is transcribed into 35S RNA, with terminal repeats, that migrates to the cytoplasm. Priming of the 35S RNA occurs by the annealing of the 3’ end of tRNA ^met to the primer-binding site (PBS) sequence, leading to the synthesis of the DNA a strand by the virus’ reverse transcriptase. Then, the RNA in the heteroduplex is degraded by the virus’ RNaseH activity, leaving purine-rich regions that act as primers for the synthesis of the positive-sense DNA b and g strands.

Our results show that AtCDA1 significantly increases the number of G to A mutations in vivo, and that there is a negative correlation between the amount of AtCDA1 mRNA present in the cell and the load reached by CaMV, suggesting that deamination of viral genomes may also constitute a significant antiviral mechanism in plants.

Lethal mutagenesis through deamination of RNA/DNA by cytidine deaminases has been proven to work as an antiviral mechanism against retroviruses ^{16– 19, 23– 27} , and some DNA ^{28– 31} and RNA ³² viruses infecting mammals. Our results show that the A. thaliana CDA1 gene has some degree of mutagenic activity on the pararetrovirus CaMV genome. Moreover, simultaneously suppressing the expression of a subset of AtCDAs, including AtCDA1, increased CaMV load, strongly suggesting an antiviral role for AtCDAs. This role of AtCDA1 is congruent with the very recent observation by Chen et al. that only the product of AtCDA1 is required for in vivo homeostasis of pyrimidines while the other eight members of the gene family may be pseudogenes ⁴⁹

Our data show that AtCDAs probably restrict CaMV replication through a process similar to the restriction of HIV-1 by APOBEC3. CaMV replicates in the cytoplasm by reverse transcription using the positive-sense 35S RNA as template. As for HIV-1, the first strand negative-sense cDNA could be deaminated during reverse transcription, transforming deoxycytidine into deoxyuridine. Then, when the positive-sense strand is produced, an A is incorporated instead of a G, increasing the proportion of G to A mutations. In the case of HIV-1, this G to A mutational bias is explained by A3G and A3H specificity for single negative stranded DNA: during HIV-1 replication, C to G transitions are rare and restricted to the PBS site and U3 regions in the 5’ long terminal repeat, where positive-stranded DNA is predicted to become transiently single stranded ⁵⁰ . Similarly, during CaMV replication the negative strand remains single stranded, while the positive is copied from it and remains double stranded ⁵¹ . Surprisingly, for AtCDA1, C to T mutations were also increased; the region studied here is close to the 5’ end of CaMV, which contains the PBS for negative-strand synthesis and the ssDNA discontinuity ∆1. The observed C to T transitions could reflect transient positive-stranded ssDNA in the 5’ terminal region during reverse transcription, nevertheless a different substrate specificity of A. thaliana CDAs cannot be ruled out.

Evidences from studies with different mammalian viruses suggest that APOBEC enzymes may have an antiviral role not only against DNA viruses and retroviruses but also against some RNA viruses ³² . Our evidences for a potential antiviral role of a plant CDA is restricted to the case of a pararetrovirus and thus the question is whether this mechanism would also operate against other types of plant viruses. Lin et al. described the spectrum of mutations accumulated in a non-coding sequence artificially inserted in the genome of turnip mosaic virus (TuMV), a prototypical RNA plant virus, during infection of N. benthamiana plants ⁵² . C to A and C to U transitions were significantly over represented in the mutant spectrum, and the authors already suggested this bias was compatible with TuMV genome being edited by CDA enzymes ⁵² .

Most of the G to A transitions detected in agroinfiltration experiments were located in the G at position 181. HIV-1 hypermutated genomes show mutational hot spots as well, which are due to preference of A3G and A3F for deamination of the third C in 5’-CCC (negative-strand) and 5’-GGC, respectively ^{53, 54} . The sequence context of the C complementary to G181 (5’-GGC) differs from what has been described for APOBEC3 as hotspot for deamination, suggesting that if AtCDAs had a context preference, it would be different from the one described for A3G. However, given the low number of mutations found, we should be cautious when concluding whether AtCDAs have a possible sequence-context preference. Since our experiments were performed in vivo, negative selection is expected to purge genomes carrying deleterious mutations. To explore this possibility, we have checked the consequence of mutations in the protein encoded by the ORF VII ( Supplementary Table S1) for the case of plant agroinfiltrated with AtCDA1 and its corresponding paired control . Eight out of the 22 different mutations observed in CaMV populations replicating in presence of AtCDA1 were nonsynonymous, thus in agreement with previous observations that most G to A transitions in CaMV are synonymous ⁵⁴ . Two remarks can be made about these numbers. First, quite surprisingly, six of these eight nonsynonymous mutations resulted in stop codons affecting two different positions (amino acids C58 and Y71). Second, transition G181A is synonymous. For CaMV replicating in the corresponding control half-leaf (agroinfiltrated with the empty pBIN61 vector), three out of the 14 different mutations observed were nonsynonymous, one of them also resulting in an stop in codon 58. No significant differences exist among the relative ratio of nonsynonymous to synonymous mutations in both samples (Fisher’s exact test P = 0.467). Despite the mutagenic effect of AtCDA1 over the CaMV population, the number of nonsynonymous mutations relative to the number of synonymous mutations is not altered, thus suggesting negative selection works, at least, as efficiently as it does in the control population. The same conclusion is reached if we only focus the comparison in the number of nonsynonymous mutations resulting in stop codons. This potential purifying effect of selection could account for our failure to detect largely hypermutated genomes, and demonstrates the need for developing new selection-free assays to further characterize AtCDA-induced mutagenesis. Despite the apparent low number of deamination mutations observed, it has a significant impact in CaMV accumulation ( Figure 2b), thus suggesting that a low threshold of G to A transition bias may be enough to lead to a reduction in viral load.

Although there is not a demonstrated correlation between the expression of APOBEC3 and mutational bias of viruses infecting mammals, caulimoviruses have an excess of G to A transitions in synonymous positions ⁵⁵ . In A. thaliana plants, we found that silencing of AtCDA1 reduced the frequency of G to A transitions in the CaMV genome, suggesting a contribution of AtCDAs to the nucleotide bias found in caulimoviruses. The increased viral load in CDA-silenced A. thaliana plants strongly suggests that deamination of viral genomes may work as an antiviral mechanism in plants, leading to questions about how general this mechanism might be, and how it may contribute to viral evolution. Describing a new natural antiviral mechanism in plants opens new research avenues for the development of new durable control strategies.

The data referenced by this article are under copyright with the following copyright statement: Copyright: � 2017 Martín S et al.

Data associated with the article are available under the terms of the Creative Commons Zero "No rights reserved" data waiver (CC0 1.0 Public domain dedication).

All datasets that support the findings in this study are available at LabArchives with DOI: 10.6070/H4TD9VD5.

‘File Sequence_data_for_Figure_1.zip’ contains the FASTA files with the sequence data used to generate the mutational spectra shown in Figure 1.

‘Data_for_Figure_2a.xlsx’ contains the AtCDA1 expression data used to generate Figure 2a.

‘Data_for_Figure_2b.xlsx’ contains the CaMV accumulation data used to generate Figure 2b.

‘Sequence_data_for_Figure_3.zip’ contains the FASTA files with sequence data used to generate the mutational spectra shown in Figure 3.