Mimivirus reveals Mre11/Rad50 fusion proteins with a sporadic distribution in eukaryotes, bacteria, viruses and plasmids

DNA double-strand breaks (DSBs) are a major cause of genomic instability and can lead to chromosomal aberration and cancer [1, 2]. DSBs occur in intermediate steps during normal DNA metabolic processes such as genome replication, meiotic recombination and programmed DNA rearrangement, but are also caused by DNA-damaging agents, including ionizing radiation as well as genotoxic chemicals. All cellular organisms possess a set of conserved proteins to cope with this dangerous form of genomic DNA; the Mre11/Rad50 complex in eukaryotes/archaea and its bacterial homologs (the SbcD/SbcC system) are the key players of the DSB metabolism generating a recombinogenic 3' overhang [3]. Due to their ubiquitous presence in cellular organisms, it has been suggested that the last universal common ancestor (LUCA) already possessed this system [4]. Mre11 and SbcD are nucleases belonging to the calcineurin-like phosphoesterase family. Rad50 and SbcC are ABC ATPases belonging to the S tructural M aintenance of C hromosomes (SMC) superfamily, and exhibit a long coiled-coil domain (~500Å when fully stretched) used to bridge two DNA molecules. The Escherichia coli SbcD/SbcC complex has an affinity for DNA hairpins and is known to generate DSBs at these sites [2, 5]. Homologous enzymes are also found in viruses; for instance, T4 phage encodes a Rad50 homolog (gp46) and a Mre11 homolog (gp47). These proteins are involved in the recombination-dependent DNA replication, an elegant solution to the end replication problem of linear viral genomes [6, 7]. In bacteria, archaea and T4, these nuclease and ATPase are encoded in an operon [2, 8].

Mimivirus (Acanthamoeba polyphaga mimivirus; APMV), a giant dsDNA virus infecting Acanthamoeba spp., is the prototype of the Mimiviridae family, the latest addition to the nucleocytoplasmic large DNA virus (NCLDV) group. Mimivirus is the largest among known viruses in both particle size (~750 nm) and genome length (1.2 Mb-genome encoding 1018 genes) [9–11]. The Mimivirus genome encodes at least eight putative DNA repair enzymes capable of correcting mismatches or errors induced by oxidation, UV irradiation and alkylating agents, including the recently analyzed putative mismatch repair enzyme MutS7 (L359) that exhibits a distinctive domain organization shared by other giant viruses [12]. Mimivirus R555 originally annotated as a Rad50 homolog [10] is part of this uniquely complete Mimivirus DNA repair tool box. The R555 gene product specifically attracted our attention following a study that pointed out the existence of numerous homologs closely related to R555 in a marine metagenomic data set [13]. Monier et al. analyzed the sequences gathered by the Global Ocean Sampling (GOS) Expedition [14], and identified 5,293 homologs for 229 Mimivirus proteins in the metagenomic data set (mostly 0.1-0.8 μm size fractions). The number of such homologs for each of the Mimivirus proteins was variable ranging from 1 to 249. R555 homologs were found to be the most abundant with 249 GOS scaffold matches, followed by 189 matches for R382 (mRNA capping enzyme) and 185 matches for R322 (B-type DNA polymerase). Here, we analyzed the sequence of R555 using database searches and phylogenetic reconstruction, and assessed the abundance of its close homologs in another large metagenomic data set generated by the Metagenomic Profiling of Nine Biomes (BIOME), which consists of 42 viral (<0.22 μm size fractions with a concentration of viral DNAs) and 45 microbial metagenomes (typically >0.22 μm size fractions) [15].

While investigating the DNA repair functions encoded by the Mimivirus genome, we discovered that the R555 initially annotated as a Rad50 homolog was in fact a fusion between Mre11 and Rad50, two proteins known to be involved in the repair of DNA double-strand breaks. Using the R555 sequence as a template, we then found that similar fusion proteins are present in a wide variety of unrelated organisms: phage, bacteria, unicellular and multicellular eukaryotes, albeit with a highly sporadic distribution. To our knowledge, this is the first description of the existence fusion genes encoding both Mre11-like and Rad50-like domains.

Interestingly, some of the fusion genes were identified in plasmids (Clostridium kluyveri and Deferribacter). In the case of Deferribacter, the plasmid-encoded sbcDC fusion gene was found flanked by insertion sequences (i.e., transposons). Our analysis of various metagenomic data sets revealed that close homologs of R555 are abundant in different aquatic environments and that they are most likely associated with viruses. Finally, our phylogenetic analysis indicated that the Mre11/Rad50 (or SbcDC) fusion proteins are only distantly related to the canonical versions of homologs and do not form a monophyletic group. No simple evolutionary scenario explains these observations. A possibility is that the cellular fusion genes were vertically derived from an ancestral gene present as a paralog of the canonical genes in an ancestral cellular organism such as LUCA. These cellular non-canonical genes were then lost from most cellular lineages but were recruited by viruses including the ancestor of Mimivirus. However, we consider this "cell-central" hypothesis unlikely as this scenario postulates numerous independent gene loss events in cellular organisms. A more likely explanation is to assume the presence of non-canonical genes in ancestral viruses and the occurrence of multiple gene fusion events.

In general, gene fusions are expected to facilitate or simplify the co-expression and assembly of protein domains initially encoded in separate genes. Such a physical link of two associated functions at the genomic level also enhances the probability of successful lateral gene transfers. Known examples of fused proteins in viruses include the primase and helicase domain fusions in large DNA viruses [23]. More recently, the mismatch repair protein MutS of Mimivirus was suggested to combine functions normally encoded in separate proteins, thanks to the fusion of the classical mismatch recognition domains with a nicking endonuclease domain [12]. The fusion proteins of this type, now classified in the MutS7 subfamily, are abundant in large DNA viruses as well as in environmental metagenomes, but also present in a few distantly related cellular organisms (i.e., the mitochondria of octocorals and Epsilonproteobacteria). Our observations on the Mre11/Rad50 fusion proteins show an intriguing resemblance to those made on the MutS7 subfamily.

The apparent contradiction between their abundance in metagenomes and the sporadic distribution in unrelated (but mostly marine) cellular organisms suggest that the true niche of these protein variants is in viruses. Viruses are abundant in aquatic environments, are known to hold a huge genetic diversity yet underrepresented in the current databases, and are suggested to be the place of the creation of new genes [24]. We propose that viruses, plasmids and/or transposons might have played a key role in the emergence of these Mre11/Rad50-like fusion proteins as well as in their subsequent propagation into different cellular organisms. The non-fused and non-canonical versions of Mre11/Rad50 homologs from viruses (such as gp46/gp47 of T4 and other marine T4-like viruses) were found outside the canonical clades corresponding to the three domains of life. We confirmed that almost all T4-like viruses with complete genomes possess homologs of these genes (data not shown). This phylogenetic feature is consistent with our hypothesis that viruses are the evolutionary origin of the Mre11/Rad50 fusion proteins found in bacteria and eukaryotes.

Gene fusions probably occurred several times in different viral lineages using an operon structure as an evolutionary template towards fused genes. Lateral transfers then possibly spread the fused genes into different viruses and cellular organisms. For instance, the presence of close homologs in mosses (Streptophyta), diatoms (stramenopiles) and green algae (Chlorophyta) suggests gene transfers among these very distantly related eukaryotes via unidentified intermediates. Forterre hypothesized that the cellular DNA informational proteins have been recruited independently in the three domains of life from different viruses, which shared a few common DNA processing enzymes such as the canonical Rad50/Mre11 [25]. In our phylogenetic tree (Figure 2(b)), cellular non-canonical versions (in either fused or non-fused forms) were found intercalated among the branches leading to non-canonical versions of viral proteins. It is possible to see this branching pattern as another (but more recent) case of the lateral flow of DNA processing genes from the virus gene pool to cellular genomes. Future efforts in generating much longer and deeper metagenomic reads are needed to better understand the evolution of these new types of proteins.

The function of R555 in the replication cycle of Mimivirus is currently unknown. In a previous work [11], R555 was found to have the same intermediary expression pattern as other genes involved in DNA replication. R555 may thus be involved in repair of transient DSBs, for instance, in the region ahead of the advancing replication fork, since the Mimivirus genome encodes a topoisomerase II (R480) [10], which removes superhelical tension by introducing transient DSBs [26]. Alternatively, the function of R555 may be associated with genomic properties more specific to Mimivirus. As mentioned earlier, the T4 proteins gp46 and gp47 (homologs of SbcC/SbcD, respectively) are involved in the recombination-dependent DNA replication (RDR) of the viral genome [6]. T4 particles carry a linear dsDNA molecule. The replication of such a linear molecule has to overcome the problem of synthesizing the complementary strand of the 3' terminus of the template DNA. T4 solves this problem using RDR linked to its terminally redundant and circularly permuted linear genome. In essence, 3' single stranded (ss) end invades homologous duplex DNA near the opposite end of the chromosome (either the same DNA molecule or its sister). gp46/gp47 appears to contribute to the generation of the 3' ss end (by a process called "resection") as well as to the bridging of two DNA ends. There is also evidence suggesting that poxviruses uses homologous recombination and DSB repair for full-size genome replication, where the G5 protein (a FEN1-like homolog) plays a crucial role [27]. The Mimivirus R555 may be involved in a similar process for the solution of the end replication problem. The chromosome of Mimivirus is not circular permuted, but exhibits a quasi-perfect (616/617 bp) inverted repetition of a 617-bp sequence near its termini [28]. These homologous regions may serve as a template to complete the replication of the ends of the Mimivirus chromosome in a way similar to the RDR used by T4 (Figure 4). The coiled-coil segment in the phage gp46 protein (homologous to SbcC) is much shorter than in its cellular counterparts. It has been suggested that this may be related to the shorter length of the T4 168 kbp genome and its much denser packing in viral replication centers compared to cellular genomes [4]. Interestingly the length of the predicted coiled-coil segment in R555 is comparable to these of cellular homologs, and appears suitable to bridge two ends of the long Mimivirus genomic DNA (Figure 1). In T4, RecA-type recombinase UvsX has a crucial role in the RDR, but Mimivirus possesses no RecA-type recombinase homolog. Mimivirus B-type DNA polymerase (R322) may be involved in such a recombination process, as the poxvirus B-type DNA polymerase is known to have a recombinase activity [29, 30].

Another plausible role of R555 in Mimivirus may be associated with the replication of DNA hairpins. In E. coli, the SbcD/SbcC complex has an affinity for DNA hairpin structures (>200 bp stem) and is known to generate DSBs at these sites, which are then repaired by homologous recombination [2, 5]. Most Mimivirus genes 3'-UTRs exhibit a palindromic sequence that serves as a polyadenylation site on the transcribed mRNA and tRNA molecules [11, 31]. These sequences have the potential of forming hairpin structures (>12 bp stem). Out of the 581 Mimivirus genes for which the 3'transcript ends were mapped, 473 (81.4%) showed these potential hairpin structures [31]. Given their large number of occurrence in the Mimivirus genome, these hairpins, albeit short, may occasionally inhibit DNA replication. R555 might thus be involved in the process ensuring the correct replication of functionally important palindromic sequences.

Homologs of the Mre11/Rad50 complex play crucial roles in the DSB repair metabolism in cellular organisms. In this study, we showed that Mimivirus R555 gene product corresponds to a fusion of Mre11-like and Rad50-like domains and that its close homologs are specifically abundant in aquatic viruses. We also identified twelve additional cases of similar fusion protein sequences in unrelated cellular organisms as well as in another virus for the first time through a systematic database search. The abundance of the Mre11/Rad50-like fusion genes in viral metagenomes and their sporadic phylogenetic distribution can be explained by recurrent creations of new variants of genes in viruses and their subsequent transfers to different cellular organisms possibly relayed by plasmids or transposons.

Mre11/Rad50 and SbcDC fusion proteins were identified by PSI-BLAST [32] using position specific scoring matrices for the COG0419 (SbcC) and COG0420 (SbcD) as queries. Prediction of coiled-coil domains was performed using the Coiled-Coil Prediction Server (http://npsa-pbil.ibcp.fr/; [33]). Multiple sequence alignments were constructed using MAFFT ver. 6 with E-INS-i option [34]. The alignments were examined and columns with more than 50% gaps were trimmed prior to phylogenetic reconstructions. Maximum likelihood phylogenetic analysis was performed using PhyML ver. 3 with the JTT substitution model and 100 bootstrap replicates [35]. We used MEGA5 (http://www.megasoftware.net/; [36]) for tree drawing. The BIOME data set was downloaded from the CAMERA web site [37]. We identified close homologs for the Mimivirus ORFs based on the following procedure [13]. First, all the Mimivirus ORF sequences were compared to the BIOME data set using TBLASTX (E-value <0.1). This initial search resulted into 13,305 metagenomic reads matching to Mimivirus ORFs. These 13,305 sequences were then searched against the UniProt database [38] using BLASTX. This search resulted into 869 metagenomic reads exhibiting their best match to Mimivirus ORFs (E-value <0.1). For each of these 869 sequences, we extracted a segment of the Mimivirus sequence that was aligned with the reads. Next, this partial Mimivirus sequence was searched against the UniProt database (excluding Mimivirus entries in the database) using BLASTP. If the best score obtained from this BLASTP search was lower than the BLASTX score obtained for the alignment of the metagenomic read and Mimivirus sequence, the metagenomic read was kept as a close homolog of the Mimivirus ORF.